Add PP-YOLOv3 code (#5281)

* [ppyolov3] add ppyolov3 base code

* add ppyolov3 s/m/x

* modify ema

* modify code to convert onnx successfully

* support arbitrary shape

* update config to use amp default

* refine ppyolo_head code

* modify reparameter code

* refine act layer

* adapter pico_head and tood_head code

* remove ppyolov3 yaml

* fix codestyle
Co-authored-by: Nwangxinxin08 <wangxinxin08@baidu.com>

ppdet/data/transform/operators.py

@@ -1747,7 +1747,7 @@ class Mixup(BaseOperator):
             gt_score2 = np.ones_like(sample[1]['gt_class'])
             gt_score = np.concatenate(
                 (gt_score1 * factor, gt_score2 * (1. - factor)), axis=0)
-            result['gt_score'] = gt_score
+            result['gt_score'] = gt_score.astype('float32')
         if 'is_crowd' in sample[0]:
             is_crowd1 = sample[0]['is_crowd']
             is_crowd2 = sample[1]['is_crowd']

ppdet/engine/callbacks.py

@@ -32,7 +32,10 @@ from ppdet.metrics import get_infer_results
 from ppdet.utils.logger import setup_logger
 logger = setup_logger('ppdet.engine')
 
-__all__ = ['Callback', 'ComposeCallback', 'LogPrinter', 'Checkpointer', 'VisualDLWriter', 'SniperProposalsGenerator']
+__all__ = [
+    'Callback', 'ComposeCallback', 'LogPrinter', 'Checkpointer',
+    'VisualDLWriter', 'SniperProposalsGenerator'
+]
 
 
 class Callback(object):
@@ -202,8 +205,14 @@ class Checkpointer(Callback):
                         logger.info("Best test {} ap is {:0.3f}.".format(
                             key, self.best_ap))
             if weight:
-                save_model(weight, self.model.optimizer, self.save_dir,
-                           save_name, epoch_id + 1)
+                if self.model.use_ema:
+                    save_model(status['weight'], self.save_dir, save_name,
+                               epoch_id + 1, self.model.optimizer)
+                    save_model(weight, self.save_dir,
+                               '{}_ema'.format(save_name), epoch_id + 1)
+                else:
+                    save_model(weight, self.save_dir, save_name, epoch_id + 1,
+                               self.model.optimizer)
 
 
 class WiferFaceEval(Callback):

ppdet/engine/trainer.py

@@ -339,7 +339,8 @@ class Trainer(object):
             self.start_epoch = load_weight(self.model.student_model, weights,
                                            self.optimizer)
         else:
-            self.start_epoch = load_weight(self.model, weights, self.optimizer)
+            self.start_epoch = load_weight(self.model, weights, self.optimizer,
+                                           self.ema if self.use_ema else None)
         logger.debug("Resume weights of epoch {}".format(self.start_epoch))
 
     def train(self, validate=False):
@@ -432,21 +433,23 @@ class Trainer(object):
                 self.status['batch_time'].update(time.time() - iter_tic)
                 self._compose_callback.on_step_end(self.status)
                 if self.use_ema:
-                    self.ema.update(self.model)
+                    self.ema.update()
                 iter_tic = time.time()
 
-            # apply ema weight on model
-            if self.use_ema:
-                weight = copy.deepcopy(self.model.state_dict())
-                self.model.set_dict(self.ema.apply())
             if self.cfg.get('unstructured_prune'):
                 self.pruner.update_params()
 
+            is_snapshot = (self._nranks < 2 or self._local_rank == 0) \
+                       and ((epoch_id + 1) % self.cfg.snapshot_epoch == 0 or epoch_id == self.end_epoch - 1)
+            if is_snapshot and self.use_ema:
+                # apply ema weight on model
+                weight = copy.deepcopy(self.model.state_dict())
+                self.model.set_dict(self.ema.apply())
+                self.status['weight'] = weight
+
             self._compose_callback.on_epoch_end(self.status)
 
-            if validate and (self._nranks < 2 or self._local_rank == 0) \
-                    and ((epoch_id + 1) % self.cfg.snapshot_epoch == 0 \
-                             or epoch_id == self.end_epoch - 1):
+            if validate and is_snapshot:
                 if not hasattr(self, '_eval_loader'):
                     # build evaluation dataset and loader
                     self._eval_dataset = self.cfg.EvalDataset
@@ -467,13 +470,15 @@ class Trainer(object):
                     Init_mark = True
                     self._init_metrics(validate=validate)
                     self._reset_metrics()
+
                 with paddle.no_grad():
                     self.status['save_best_model'] = True
                     self._eval_with_loader(self._eval_loader)
 
-            # restore origin weight on model
-            if self.use_ema:
+            if is_snapshot and self.use_ema:
+                # reset original weight
                 self.model.set_dict(weight)
+                self.status.pop('weight')
 
         self._compose_callback.on_train_end(self.status)
 
@@ -634,6 +639,11 @@ class Trainer(object):
 
         if hasattr(self.model, 'deploy'):
             self.model.deploy = True
+
+        for layer in self.model.sublayers():
+            if hasattr(layer, 'convert_to_deploy'):
+                layer.convert_to_deploy()
+
         export_post_process = self.cfg.get('export_post_process', False)
         if hasattr(self.model, 'export_post_process'):
             self.model.export_post_process = export_post_process

ppdet/modeling/architectures/yolo.py

@@ -109,10 +109,14 @@ class YOLOv3(BaseArch):
                 if self.return_idx:
                     _, bbox, bbox_num, _ = self.post_process(
                         yolo_head_outs, self.yolo_head.mask_anchors)
-                else:
+                elif self.post_process is not None:
                     bbox, bbox_num = self.post_process(
                         yolo_head_outs, self.yolo_head.mask_anchors,
                         self.inputs['im_shape'], self.inputs['scale_factor'])
+                else:
+                    bbox, bbox_num = self.yolo_head.post_process(
+                        yolo_head_outs, self.inputs['im_shape'],
+                        self.inputs['scale_factor'])
                 output = {'bbox': bbox, 'bbox_num': bbox_num}
 
             return output

ppdet/modeling/assigners/utils.py

@@ -107,11 +107,15 @@ def gather_topk_anchors(metrics, topk, largest=True, topk_mask=None, eps=1e-9):
     return is_in_topk.astype(metrics.dtype)
 
 
-def check_points_inside_bboxes(points, bboxes, eps=1e-9):
+def check_points_inside_bboxes(points,
+                               bboxes,
+                               center_radius_tensor=None,
+                               eps=1e-9):
     r"""
     Args:
         points (Tensor, float32): shape[L, 2], "xy" format, L: num_anchors
         bboxes (Tensor, float32): shape[B, n, 4], "xmin, ymin, xmax, ymax" format
+        center_radius_tensor (Tensor, float32): shape [L, 1] Default: None.
         eps (float): Default: 1e-9
     Returns:
         is_in_bboxes (Tensor, float32): shape[B, n, L], value=1. means selected
@@ -119,6 +123,19 @@ def check_points_inside_bboxes(points, bboxes, eps=1e-9):
     points = points.unsqueeze([0, 1])
     x, y = points.chunk(2, axis=-1)
     xmin, ymin, xmax, ymax = bboxes.unsqueeze(2).chunk(4, axis=-1)
+    if center_radius_tensor is not None:
+        center_radius_tensor = center_radius_tensor.unsqueeze([0, 1])
+        bboxes_cx = (xmin + xmax) / 2.
+        bboxes_cy = (ymin + ymax) / 2.
+        xmin_sampling = bboxes_cx - center_radius_tensor
+        ymin_sampling = bboxes_cy - center_radius_tensor
+        xmax_sampling = bboxes_cx + center_radius_tensor
+        ymax_sampling = bboxes_cy + center_radius_tensor
+
+        xmin = paddle.maximum(xmin, xmin_sampling)
+        ymin = paddle.maximum(ymin, ymin_sampling)
+        xmax = paddle.minimum(xmax, xmax_sampling)
+        ymax = paddle.minimum(ymax, ymax_sampling)
     l = x - xmin
     t = y - ymin
     r = xmax - x
@@ -167,14 +184,16 @@ def generate_anchors_for_grid_cell(feats,
         grid_cell_size (float): anchor size
         grid_cell_offset (float): The range is between 0 and 1.
     Returns:
-        anchors (List[Tensor]): shape[s, (l, 4)]
-        num_anchors_list (List[int]): shape[s]
-        stride_tensor_list (List[Tensor]): shape[s, (l, 1)]
+        anchors (Tensor): shape[l, 4], "xmin, ymin, xmax, ymax" format.
+        anchor_points (Tensor): shape[l, 2], "x, y" format.
+        num_anchors_list (List[int]): shape[s], contains [s_1, s_2, ...].
+        stride_tensor (Tensor): shape[l, 1], contains the stride for each scale.
     """
     assert len(feats) == len(fpn_strides)
     anchors = []
+    anchor_points = []
     num_anchors_list = []
-    stride_tensor_list = []
+    stride_tensor = []
     for feat, stride in zip(feats, fpn_strides):
         _, _, h, w = feat.shape
         cell_half_size = grid_cell_size * stride * 0.5
@@ -187,8 +206,19 @@ def generate_anchors_for_grid_cell(feats,
                 shift_x + cell_half_size, shift_y + cell_half_size
             ],
             axis=-1).astype(feat.dtype)
+        anchor_point = paddle.stack(
+            [shift_x, shift_y], axis=-1).astype(feat.dtype)
+
         anchors.append(anchor.reshape([-1, 4]))
+        anchor_points.append(anchor_point.reshape([-1, 2]))
         num_anchors_list.append(len(anchors[-1]))
-        stride_tensor_list.append(
-            paddle.full([num_anchors_list[-1], 1], stride))
-    return anchors, num_anchors_list, stride_tensor_list
+        stride_tensor.append(
+            paddle.full(
+                [num_anchors_list[-1], 1], stride, dtype=feat.dtype))
+    anchors = paddle.concat(anchors)
+    anchors.stop_gradient = True
+    anchor_points = paddle.concat(anchor_points)
+    anchor_points.stop_gradient = True
+    stride_tensor = paddle.concat(stride_tensor)
+    stride_tensor.stop_gradient = True
+    return anchors, anchor_points, num_anchors_list, stride_tensor

ppdet/modeling/backbones/__init__.py

@@ -29,6 +29,7 @@ from . import swin_transformer
 from . import lcnet
 from . import hardnet
 from . import esnet
+from . import cspresnet
 
 from .vgg import *
 from .resnet import *
@@ -47,3 +48,4 @@ from .swin_transformer import *
 from .lcnet import *
 from .hardnet import *
 from .esnet import *
+from .cspresnet import *

ppdet/modeling/backbones/cspresnet.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+from paddle.regularizer import L2Decay
+
+from ppdet.modeling.ops import get_act_fn
+from ppdet.core.workspace import register, serializable
+from ..shape_spec import ShapeSpec
+
+__all__ = ['CSPResNet', 'BasicBlock', 'EffectiveSELayer', 'ConvBNLayer']
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(self,
+                 ch_in,
+                 ch_out,
+                 filter_size=3,
+                 stride=1,
+                 groups=1,
+                 padding=0,
+                 act=None):
+        super(ConvBNLayer, self).__init__()
+
+        self.conv = nn.Conv2D(
+            in_channels=ch_in,
+            out_channels=ch_out,
+            kernel_size=filter_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            bias_attr=False)
+
+        self.bn = nn.BatchNorm2D(
+            ch_out,
+            weight_attr=ParamAttr(regularizer=L2Decay(0.0)),
+            bias_attr=ParamAttr(regularizer=L2Decay(0.0)))
+        self.act = get_act_fn(act) if act is None or isinstance(act, (
+            str, dict)) else act
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        x = self.act(x)
+
+        return x
+
+
+class RepVggBlock(nn.Layer):
+    def __init__(self, ch_in, ch_out, act='relu'):
+        super(RepVggBlock, self).__init__()
+        self.ch_in = ch_in
+        self.ch_out = ch_out
+        self.conv1 = ConvBNLayer(
+            ch_in, ch_out, 3, stride=1, padding=1, act=None)
+        self.conv2 = ConvBNLayer(
+            ch_in, ch_out, 1, stride=1, padding=0, act=None)
+        self.act = get_act_fn(act) if act is None or isinstance(act, (
+            str, dict)) else act
+
+    def forward(self, x):
+        if hasattr(self, 'conv'):
+            y = self.conv(x)
+        else:
+            y = self.conv1(x) + self.conv2(x)
+        y = self.act(y)
+        return y
+
+    def convert_to_deploy(self):
+        if not hasattr(self, 'conv'):
+            self.conv = nn.Conv2D(
+                in_channels=self.ch_in,
+                out_channels=self.ch_out,
+                kernel_size=3,
+                stride=1,
+                padding=1,
+                groups=1)
+        kernel, bias = self.get_equivalent_kernel_bias()
+        self.conv.weight.set_value(kernel)
+        self.conv.bias.set_value(bias)
+
+    def get_equivalent_kernel_bias(self):
+        kernel3x3, bias3x3 = self._fuse_bn_tensor(self.conv1)
+        kernel1x1, bias1x1 = self._fuse_bn_tensor(self.conv2)
+        return kernel3x3 + self._pad_1x1_to_3x3_tensor(
+            kernel1x1), bias3x3 + bias1x1
+
+    def _pad_1x1_to_3x3_tensor(self, kernel1x1):
+        if kernel1x1 is None:
+            return 0
+        else:
+            return nn.functional.pad(kernel1x1, [1, 1, 1, 1])
+
+    def _fuse_bn_tensor(self, branch):
+        if branch is None:
+            return 0, 0
+        kernel = branch.conv.weight
+        running_mean = branch.bn._mean
+        running_var = branch.bn._variance
+        gamma = branch.bn.weight
+        beta = branch.bn.bias
+        eps = branch.bn._epsilon
+        std = (running_var + eps).sqrt()
+        t = (gamma / std).reshape((-1, 1, 1, 1))
+        return kernel * t, beta - running_mean * gamma / std
+
+
+class BasicBlock(nn.Layer):
+    def __init__(self, ch_in, ch_out, act='relu', shortcut=True):
+        super(BasicBlock, self).__init__()
+        assert ch_in == ch_out
+        self.conv1 = ConvBNLayer(ch_in, ch_out, 3, stride=1, padding=1, act=act)
+        self.conv2 = RepVggBlock(ch_out, ch_out, act=act)
+        self.shortcut = shortcut
+
+    def forward(self, x):
+        y = self.conv1(x)
+        y = self.conv2(y)
+        if self.shortcut:
+            return paddle.add(x, y)
+        else:
+            return y
+
+
+class EffectiveSELayer(nn.Layer):
+    """ Effective Squeeze-Excitation
+    From `CenterMask : Real-Time Anchor-Free Instance Segmentation` - https://arxiv.org/abs/1911.06667
+    """
+
+    def __init__(self, channels, act='hardsigmoid'):
+        super(EffectiveSELayer, self).__init__()
+        self.fc = nn.Conv2D(channels, channels, kernel_size=1, padding=0)
+        self.act = get_act_fn(act) if act is None or isinstance(act, (
+            str, dict)) else act
+
+    def forward(self, x):
+        x_se = x.mean((2, 3), keepdim=True)
+        x_se = self.fc(x_se)
+        return x * self.act(x_se)
+
+
+class CSPResStage(nn.Layer):
+    def __init__(self,
+                 block_fn,
+                 ch_in,
+                 ch_out,
+                 n,
+                 stride,
+                 act='relu',
+                 attn='eca'):
+        super(CSPResStage, self).__init__()
+
+        ch_mid = (ch_in + ch_out) // 2
+        if stride == 2:
+            self.conv_down = ConvBNLayer(
+                ch_in, ch_mid, 3, stride=2, padding=1, act=act)
+        else:
+            self.conv_down = None
+        self.conv1 = ConvBNLayer(ch_mid, ch_mid // 2, 1, act=act)
+        self.conv2 = ConvBNLayer(ch_mid, ch_mid // 2, 1, act=act)
+        self.blocks = nn.Sequential(* [
+            block_fn(
+                ch_mid // 2, ch_mid // 2, act=act, shortcut=True)
+            for i in range(n)
+        ])
+        if attn:
+            self.attn = EffectiveSELayer(ch_mid, act='hardsigmoid')
+        else:
+            self.attn = None
+
+        self.conv3 = ConvBNLayer(ch_mid, ch_out, 1, act=act)
+
+    def forward(self, x):
+        if self.conv_down is not None:
+            x = self.conv_down(x)
+        y1 = self.conv1(x)
+        y2 = self.blocks(self.conv2(x))
+        y = paddle.concat([y1, y2], axis=1)
+        if self.attn is not None:
+            y = self.attn(y)
+        y = self.conv3(y)
+        return y
+
+
+@register
+@serializable
+class CSPResNet(nn.Layer):
+    __shared__ = ['width_mult', 'depth_mult', 'trt']
+
+    def __init__(self,
+                 layers=[3, 6, 6, 3],
+                 channels=[64, 128, 256, 512, 1024],
+                 act='swish',
+                 return_idx=[0, 1, 2, 3, 4],
+                 depth_wise=False,
+                 use_large_stem=False,
+                 width_mult=1.0,
+                 depth_mult=1.0,
+                 trt=False):
+        super(CSPResNet, self).__init__()
+        channels = [max(round(c * width_mult), 1) for c in channels]
+        layers = [max(round(l * depth_mult), 1) for l in layers]
+        act = get_act_fn(
+            act, trt=trt) if act is None or isinstance(act,
+                                                       (str, dict)) else act
+
+        if use_large_stem:
+            self.stem = nn.Sequential(
+                ('conv1', ConvBNLayer(
+                    3, channels[0] // 2, 3, stride=2, padding=1, act=act)),
+                ('conv2', ConvBNLayer(
+                    channels[0] // 2,
+                    channels[0] // 2,
+                    3,
+                    stride=1,
+                    padding=1,
+                    act=act)), ('conv3', ConvBNLayer(
+                        channels[0] // 2,
+                        channels[0],
+                        3,
+                        stride=1,
+                        padding=1,
+                        act=act)))
+        else:
+            self.stem = nn.Sequential(
+                ('conv1', ConvBNLayer(
+                    3, channels[0] // 2, 3, stride=2, padding=1, act=act)),
+                ('conv2', ConvBNLayer(
+                    channels[0] // 2,
+                    channels[0],
+                    3,
+                    stride=1,
+                    padding=1,
+                    act=act)))
+
+        n = len(channels) - 1
+        self.stages = nn.Sequential(* [(str(i), CSPResStage(
+            BasicBlock, channels[i], channels[i + 1], layers[i], 2, act=act))
+                                       for i in range(n)])
+
+        self._out_channels = channels[1:]
+        self._out_strides = [4, 8, 16, 32]
+        self.return_idx = return_idx
+
+    def forward(self, inputs):
+        x = inputs['image']
+        x = self.stem(x)
+        outs = []
+        for idx, stage in enumerate(self.stages):
+            x = stage(x)
+            if idx in self.return_idx:
+                outs.append(x)
+
+        return outs
+
+    @property
+    def out_shape(self):
+        return [
+            ShapeSpec(
+                channels=self._out_channels[i], stride=self._out_strides[i])
+            for i in self.return_idx
+        ]

ppdet/modeling/backbones/darknet.py

@@ -77,8 +77,8 @@ class ConvBNLayer(nn.Layer):
         out = self.batch_norm(out)
         if self.act == 'leaky':
             out = F.leaky_relu(out, 0.1)
-        elif self.act == 'mish':
-            out = mish(out)
+        else:
+            out = getattr(F, self.act)(out)
         return out
 
 
@@ -156,7 +156,7 @@ class BasicBlock(nn.Layer):
         # channel route: 10-->5 --> 5-->10
         self.conv1 = ConvBNLayer(
             ch_in=ch_in,
-            ch_out=int(ch_out/2),
+            ch_out=int(ch_out / 2),
             filter_size=1,
             stride=1,
             padding=0,
@@ -165,8 +165,8 @@ class BasicBlock(nn.Layer):
             freeze_norm=freeze_norm,
             data_format=data_format)
         self.conv2 = ConvBNLayer(
-            ch_in=int(ch_out/2),
-            ch_out=ch_out ,
+            ch_in=int(ch_out / 2),
+            ch_out=ch_out,
             filter_size=3,
             stride=1,
             padding=1,
@@ -317,7 +317,7 @@ class DarkNet(nn.Layer):
                 down_name,
                 DownSample(
                     ch_in=int(ch_in[i]),
-                    ch_out=int(ch_in[i+1]),
+                    ch_out=int(ch_in[i + 1]),
                     norm_type=norm_type,
                     norm_decay=norm_decay,
                     freeze_norm=freeze_norm,

ppdet/modeling/bbox_utils.py

@@ -744,9 +744,9 @@ def distance2bbox(points, distance, max_shape=None):
 def bbox_center(boxes):
     """Get bbox centers from boxes.
     Args:
-        boxes (Tensor): boxes with shape (N, 4), "xmin, ymin, xmax, ymax" format.
+        boxes (Tensor): boxes with shape (..., 4), "xmin, ymin, xmax, ymax" format.
     Returns:
-        Tensor: boxes centers with shape (N, 2), "cx, cy" format.
+        Tensor: boxes centers with shape (..., 2), "cx, cy" format.
     """
     boxes_cx = (boxes[..., 0] + boxes[..., 2]) / 2
     boxes_cy = (boxes[..., 1] + boxes[..., 3]) / 2
@@ -782,7 +782,7 @@ def delta2bbox_v2(rois,
                   means=(0.0, 0.0, 0.0, 0.0),
                   stds=(1.0, 1.0, 1.0, 1.0),
                   max_shape=None,
-                  wh_ratio_clip=16.0/1000.0,
+                  wh_ratio_clip=16.0 / 1000.0,
                   ctr_clip=None):
     """Transform network output(delta) to bboxes.
     Based on https://github.com/open-mmlab/mmdetection/blob/master/mmdet/core/
@@ -860,7 +860,7 @@ def bbox2delta_v2(src_boxes,
     dw = paddle.log(tgt_w / src_w)
     dh = paddle.log(tgt_h / src_h)
 
-    deltas = paddle.stack((dx, dy, dw, dh), axis=1) # [n, 4]
+    deltas = paddle.stack((dx, dy, dw, dh), axis=1)  # [n, 4]
     means = paddle.to_tensor(means, place=src_boxes.place)
     stds = paddle.to_tensor(stds, place=src_boxes.place)
     deltas = (deltas - means) / stds

ppdet/modeling/heads/__init__.py

@@ -32,6 +32,7 @@ from . import detr_head
 from . import sparsercnn_head
 from . import tood_head
 from . import retina_head
+from . import ppyolo_head
 
 from .bbox_head import *
 from .mask_head import *
@@ -53,3 +54,4 @@ from .detr_head import *
 from .sparsercnn_head import *
 from .tood_head import *
 from .retina_head import *
+from .ppyolo_head import *

ppdet/modeling/heads/pico_head.py

@@ -464,12 +464,13 @@ class PicoHeadV2(GFLHead):
         assert len(fpn_feats) == len(
             self.fpn_stride
         ), "The size of fpn_feats is not equal to size of fpn_stride"
-        anchors, num_anchors_list, stride_tensor_list = generate_anchors_for_grid_cell(
+        anchors, _, num_anchors_list, stride_tensor_list = generate_anchors_for_grid_cell(
             fpn_feats, self.fpn_stride, self.grid_cell_scale, self.cell_offset)
+        anchors_split = paddle.split(anchors, num_anchors_list)
 
         cls_score_list, reg_list, box_list = [], [], []
         for i, fpn_feat, anchor, stride, align_cls in zip(
-                range(len(self.fpn_stride)), fpn_feats, anchors,
+                range(len(self.fpn_stride)), fpn_feats, anchors_split,
                 self.fpn_stride, self.cls_align):
             b, _, h, w = get_static_shape(fpn_feat)
             # task decomposition
@@ -507,10 +508,6 @@ class PicoHeadV2(GFLHead):
             cls_score_list = paddle.concat(cls_score_list, axis=1)
             box_list = paddle.concat(box_list, axis=1)
             reg_list = paddle.concat(reg_list, axis=1)
-            anchors = paddle.concat(anchors)
-            anchors.stop_gradient = True
-            stride_tensor_list = paddle.concat(stride_tensor_list)
-            stride_tensor_list.stop_gradient = True
             return cls_score_list, reg_list, box_list, anchors, num_anchors_list, stride_tensor_list
 
     def get_loss(self, head_outs, gt_meta):

ppdet/modeling/heads/ppyolo_head.py

+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from ppdet.core.workspace import register
+
+from ..bbox_utils import batch_distance2bbox
+from ..losses import GIoULoss
+from ..initializer import bias_init_with_prob, constant_, normal_
+from ..assigners.utils import generate_anchors_for_grid_cell
+from ppdet.modeling.backbones.cspresnet import ConvBNLayer
+from ppdet.modeling.ops import get_static_shape, paddle_distributed_is_initialized, get_act_fn
+
+__all__ = ['PPYOLOHead']
+
+
+class ESEAttn(nn.Layer):
+    def __init__(self, feat_channels, act='swish'):
+        super(ESEAttn, self).__init__()
+        self.fc = nn.Conv2D(feat_channels, feat_channels, 1)
+        self.conv = ConvBNLayer(feat_channels, feat_channels, 1, act=act)
+
+        self._init_weights()
+
+    def _init_weights(self):
+        normal_(self.fc.weight, std=0.001)
+
+    def forward(self, feat, avg_feat):
+        weight = F.sigmoid(self.fc(avg_feat))
+        return self.conv(feat * weight)
+
+
+@register
+class PPYOLOHead(nn.Layer):
+    __shared__ = ['num_classes', 'trt']
+    __inject__ = ['static_assigner', 'assigner', 'nms']
+
+    def __init__(self,
+                 in_channels=[1024, 512, 256],
+                 num_classes=80,
+                 act='swish',
+                 fpn_strides=(32, 16, 8),
+                 grid_cell_scale=5.0,
+                 grid_cell_offset=0.5,
+                 reg_max=16,
+                 static_assigner_epoch=4,
+                 use_varifocal_loss=True,
+                 static_assigner='ATSSAssigner',
+                 assigner='TaskAlignedAssigner',
+                 nms='MultiClassNMS',
+                 eval_input_size=[],
+                 loss_weight={
+                     'class': 1.0,
+                     'iou': 2.5,
+                     'dfl': 0.5,
+                 },
+                 trt=False):
+        super(PPYOLOHead, self).__init__()
+        assert len(in_channels) > 0, "len(in_channels) should > 0"
+        self.in_channels = in_channels
+        self.num_classes = num_classes
+        self.fpn_strides = fpn_strides
+        self.grid_cell_scale = grid_cell_scale
+        self.grid_cell_offset = grid_cell_offset
+        self.reg_max = reg_max
+        self.iou_loss = GIoULoss()
+        self.loss_weight = loss_weight
+        self.use_varifocal_loss = use_varifocal_loss
+        self.eval_input_size = eval_input_size
+
+        self.static_assigner_epoch = static_assigner_epoch
+        self.static_assigner = static_assigner
+        self.assigner = assigner
+        self.nms = nms
+        # stem
+        self.stem_cls = nn.LayerList()
+        self.stem_reg = nn.LayerList()
+        act = get_act_fn(
+            act, trt=trt) if act is None or isinstance(act,
+                                                       (str, dict)) else act
+        for in_c in self.in_channels:
+            self.stem_cls.append(ESEAttn(in_c, act=act))
+            self.stem_reg.append(ESEAttn(in_c, act=act))
+        # pred head
+        self.pred_cls = nn.LayerList()
+        self.pred_reg = nn.LayerList()
+        for in_c in self.in_channels:
+            self.pred_cls.append(
+                nn.Conv2D(
+                    in_c, self.num_classes, 3, padding=1))
+            self.pred_reg.append(
+                nn.Conv2D(
+                    in_c, 4 * (self.reg_max + 1), 3, padding=1))
+        # projection conv
+        self.proj_conv = nn.Conv2D(self.reg_max + 1, 1, 1, bias_attr=False)
+        self._init_weights()
+
+    @classmethod
+    def from_config(cls, cfg, input_shape):
+        return {'in_channels': [i.channels for i in input_shape], }
+
+    def _init_weights(self):
+        bias_cls = bias_init_with_prob(0.01)
+        for cls_, reg_ in zip(self.pred_cls, self.pred_reg):
+            constant_(cls_.weight)
+            constant_(cls_.bias, bias_cls)
+            constant_(reg_.weight)
+            constant_(reg_.bias, 1.0)
+
+        self.proj = paddle.linspace(0, self.reg_max, self.reg_max + 1)
+        self.proj_conv.weight.set_value(
+            self.proj.reshape([1, self.reg_max + 1, 1, 1]))
+        self.proj_conv.weight.stop_gradient = True
+
+        if self.eval_input_size:
+            anchor_points, stride_tensor = self._generate_anchors()
+            self.register_buffer('anchor_points', anchor_points)
+            self.register_buffer('stride_tensor', stride_tensor)
+
+    def forward_train(self, feats, targets):
+        anchors, anchor_points, num_anchors_list, stride_tensor = \
+            generate_anchors_for_grid_cell(
+                feats, self.fpn_strides, self.grid_cell_scale,
+                self.grid_cell_offset)
+
+        cls_score_list, reg_distri_list = [], []
+        for i, feat in enumerate(feats):
+            avg_feat = F.adaptive_avg_pool2d(feat, (1, 1))
+            cls_logit = self.pred_cls[i](self.stem_cls[i](feat, avg_feat) +
+                                         feat)
+            reg_distri = self.pred_reg[i](self.stem_reg[i](feat, avg_feat))
+            # cls and reg
+            cls_score = F.sigmoid(cls_logit)
+            cls_score_list.append(cls_score.flatten(2).transpose([0, 2, 1]))
+            reg_distri_list.append(reg_distri.flatten(2).transpose([0, 2, 1]))
+        cls_score_list = paddle.concat(cls_score_list, axis=1)
+        reg_distri_list = paddle.concat(reg_distri_list, axis=1)
+
+        return self.get_loss([
+            cls_score_list, reg_distri_list, anchors, anchor_points,
+            num_anchors_list, stride_tensor
+        ], targets)
+
+    def _generate_anchors(self, feats=None):
+        # just use in eval time
+        anchor_points = []
+        stride_tensor = []
+        for i, stride in enumerate(self.fpn_strides):
+            if feats is not None:
+                _, _, h, w = feats[i].shape
+            else:
+                h = int(self.eval_input_size[0] / stride)
+                w = int(self.eval_input_size[1] / stride)
+            shift_x = paddle.arange(end=w) + self.grid_cell_offset
+            shift_y = paddle.arange(end=h) + self.grid_cell_offset
+            shift_y, shift_x = paddle.meshgrid(shift_y, shift_x)
+            anchor_point = paddle.cast(
+                paddle.stack(
+                    [shift_x, shift_y], axis=-1), dtype='float32')
+            anchor_points.append(anchor_point.reshape([-1, 2]))
+            stride_tensor.append(
+                paddle.full(
+                    [h * w, 1], stride, dtype='float32'))
+        anchor_points = paddle.concat(anchor_points)
+        stride_tensor = paddle.concat(stride_tensor)
+        return anchor_points, stride_tensor
+
+    def forward_eval(self, feats):
+        if self.eval_input_size:
+            anchor_points, stride_tensor = self.anchor_points, self.stride_tensor
+        else:
+            anchor_points, stride_tensor = self._generate_anchors(feats)
+        cls_score_list, reg_dist_list = [], []
+        for i, feat in enumerate(feats):
+            b, _, h, w = feat.shape
+            l = h * w
+            avg_feat = F.adaptive_avg_pool2d(feat, (1, 1))
+            cls_logit = self.pred_cls[i](self.stem_cls[i](feat, avg_feat) +
+                                         feat)
+            reg_dist = self.pred_reg[i](self.stem_reg[i](feat, avg_feat))
+            reg_dist = reg_dist.reshape([-1, 4, self.reg_max + 1, l]).transpose(
+                [0, 2, 1, 3])
+            reg_dist = self.proj_conv(F.softmax(reg_dist, axis=1))
+            # cls and reg
+            cls_score = F.sigmoid(cls_logit)
+            cls_score_list.append(cls_score.reshape([b, self.num_classes, l]))
+            reg_dist_list.append(reg_dist.reshape([b, 4, l]))
+
+        cls_score_list = paddle.concat(cls_score_list, axis=-1)
+        reg_dist_list = paddle.concat(reg_dist_list, axis=-1)
+
+        return cls_score_list, reg_dist_list, anchor_points, stride_tensor
+
+    def forward(self, feats, targets=None):
+        assert len(feats) == len(self.fpn_strides), \
+            "The size of feats is not equal to size of fpn_strides"
+
+        if self.training:
+            return self.forward_train(feats, targets)
+        else:
+            return self.forward_eval(feats)
+
+    @staticmethod
+    def _focal_loss(score, label, alpha=0.25, gamma=2.0):
+        weight = (score - label).pow(gamma)
+        if alpha > 0:
+            alpha_t = alpha * label + (1 - alpha) * (1 - label)
+            weight *= alpha_t
+        loss = F.binary_cross_entropy(
+            score, label, weight=weight, reduction='sum')
+        return loss
+
+    @staticmethod
+    def _varifocal_loss(pred_score, gt_score, label, alpha=0.75, gamma=2.0):
+        weight = alpha * pred_score.pow(gamma) * (1 - label) + gt_score * label
+        loss = F.binary_cross_entropy(
+            pred_score, gt_score, weight=weight, reduction='sum')
+        return loss
+
+    def _bbox_decode(self, anchor_points, pred_dist):
+        b, l, _ = get_static_shape(pred_dist)
+        pred_dist = F.softmax(pred_dist.reshape([b, l, 4, self.reg_max + 1
+                                                 ])).matmul(self.proj)
+        return batch_distance2bbox(anchor_points, pred_dist)
+
+    def _bbox2distance(self, points, bbox):
+        x1y1, x2y2 = paddle.split(bbox, 2, -1)
+        lt = points - x1y1
+        rb = x2y2 - points
+        return paddle.concat([lt, rb], -1).clip(0, self.reg_max - 0.01)
+
+    def _df_loss(self, pred_dist, target):
+        target_left = paddle.cast(target, 'int64')
+        target_right = target_left + 1
+        weight_left = target_right.astype('float32') - target
+        weight_right = 1 - weight_left
+        loss_left = F.cross_entropy(
+            pred_dist, target_left, reduction='none') * weight_left
+        loss_right = F.cross_entropy(
+            pred_dist, target_right, reduction='none') * weight_right
+        return (loss_left + loss_right).mean(-1, keepdim=True)
+
+    def _bbox_loss(self, pred_dist, pred_bboxes, anchor_points, assigned_labels,
+                   assigned_bboxes, assigned_scores, assigned_scores_sum):
+        # select positive samples mask
+        mask_positive = (assigned_labels != self.num_classes)
+        num_pos = mask_positive.sum()
+        # pos/neg loss
+        if num_pos > 0:
+            # l1 + iou
+            bbox_mask = mask_positive.unsqueeze(-1).tile([1, 1, 4])
+            pred_bboxes_pos = paddle.masked_select(pred_bboxes,
+                                                   bbox_mask).reshape([-1, 4])
+            assigned_bboxes_pos = paddle.masked_select(
+                assigned_bboxes, bbox_mask).reshape([-1, 4])
+            bbox_weight = paddle.masked_select(
+                assigned_scores.sum(-1), mask_positive).unsqueeze(-1)
+
+            loss_l1 = F.l1_loss(pred_bboxes_pos, assigned_bboxes_pos)
+
+            loss_iou = self.iou_loss(pred_bboxes_pos,
+                                     assigned_bboxes_pos) * bbox_weight
+            loss_iou = loss_iou.sum() / assigned_scores_sum
+
+            dist_mask = mask_positive.unsqueeze(-1).tile(
+                [1, 1, (self.reg_max + 1) * 4])
+            pred_dist_pos = paddle.masked_select(
+                pred_dist, dist_mask).reshape([-1, 4, self.reg_max + 1])
+            assigned_ltrb = self._bbox2distance(anchor_points, assigned_bboxes)
+            assigned_ltrb_pos = paddle.masked_select(
+                assigned_ltrb, bbox_mask).reshape([-1, 4])
+            loss_dfl = self._df_loss(pred_dist_pos,
+                                     assigned_ltrb_pos) * bbox_weight
+            loss_dfl = loss_dfl.sum() / assigned_scores_sum
+        else:
+            loss_l1 = paddle.zeros([1])
+            loss_iou = paddle.zeros([1])
+            loss_dfl = paddle.zeros([1])
+        return loss_l1, loss_iou, loss_dfl
+
+    def get_loss(self, head_outs, gt_meta):
+        pred_scores, pred_distri, anchors,\
+        anchor_points, num_anchors_list, stride_tensor = head_outs
+
+        anchor_points_s = anchor_points / stride_tensor
+        pred_bboxes = self._bbox_decode(anchor_points_s, pred_distri)
+
+        gt_labels = gt_meta['gt_class']
+        gt_bboxes = gt_meta['gt_bbox']
+        pad_gt_mask = gt_meta['pad_gt_mask']
+        # label assignment
+        if gt_meta['epoch_id'] < self.static_assigner_epoch:
+            assigned_labels, assigned_bboxes, assigned_scores = \
+                self.static_assigner(
+                    anchors,
+                    num_anchors_list,
+                    gt_labels,
+                    gt_bboxes,
+                    pad_gt_mask,
+                    bg_index=self.num_classes,
+                    pred_bboxes=pred_bboxes.detach() * stride_tensor)
+            alpha_l = 0.25
+        else:
+            assigned_labels, assigned_bboxes, assigned_scores = \
+                self.assigner(
+                pred_scores.detach(),
+                pred_bboxes.detach() * stride_tensor,
+                anchor_points,
+                num_anchors_list,
+                gt_labels,
+                gt_bboxes,
+                pad_gt_mask,
+                bg_index=self.num_classes)
+            alpha_l = -1
+        # rescale bbox
+        assigned_bboxes /= stride_tensor
+        # cls loss
+        if self.use_varifocal_loss:
+            one_hot_label = F.one_hot(assigned_labels, self.num_classes)
+            loss_cls = self._varifocal_loss(pred_scores, assigned_scores,
+                                            one_hot_label)
+        else:
+            loss_cls = self._focal_loss(
+                pred_scores, assigned_scores, alpha=alpha_l)
+
+        assigned_scores_sum = assigned_scores.sum()
+        if paddle_distributed_is_initialized():
+            paddle.distributed.all_reduce(assigned_scores_sum)
+            assigned_scores_sum = paddle.clip(
+                assigned_scores_sum / paddle.distributed.get_world_size(),
+                min=1)
+        loss_cls /= assigned_scores_sum
+
+        loss_l1, loss_iou, loss_dfl = \
+            self._bbox_loss(pred_distri, pred_bboxes, anchor_points_s,
+                            assigned_labels, assigned_bboxes, assigned_scores,
+                            assigned_scores_sum)
+        loss = self.loss_weight['class'] * loss_cls + \
+               self.loss_weight['iou'] * loss_iou + \
+               self.loss_weight['dfl'] * loss_dfl
+        out_dict = {
+            'loss': loss,
+            'loss_cls': loss_cls,
+            'loss_iou': loss_iou,
+            'loss_dfl': loss_dfl,
+            'loss_l1': loss_l1,
+        }
+        return out_dict
+
+    def post_process(self, head_outs, img_shape, scale_factor):
+        pred_scores, pred_dist, anchor_points, stride_tensor = head_outs
+        pred_bboxes = batch_distance2bbox(anchor_points,
+                                          pred_dist.transpose([0, 2, 1]))
+        pred_bboxes *= stride_tensor
+        # scale bbox to origin
+        scale_y, scale_x = paddle.split(scale_factor, 2, axis=-1)
+        scale_factor = paddle.concat(
+            [scale_x, scale_y, scale_x, scale_y], axis=-1).reshape([-1, 1, 4])
+        pred_bboxes /= scale_factor
+        bbox_pred, bbox_num, _ = self.nms(pred_bboxes, pred_scores)
+        return bbox_pred, bbox_num

ppdet/modeling/heads/tood_head.py

@@ -218,13 +218,17 @@ class TOODHead(nn.Layer):
         assert len(feats) == len(self.fpn_strides), \
             "The size of feats is not equal to size of fpn_strides"
 
-        anchors, num_anchors_list, stride_tensor_list = generate_anchors_for_grid_cell(
+        anchors, anchor_points, num_anchors_list, stride_tensor =\
+            generate_anchors_for_grid_cell(
             feats, self.fpn_strides, self.grid_cell_scale,
             self.grid_cell_offset)
+        anchor_centers_split = paddle.split(anchor_points / stride_tensor,
+                                            num_anchors_list)
 
         cls_score_list, bbox_pred_list = [], []
-        for feat, scale_reg, anchor, stride in zip(feats, self.scales_regs,
-                                                   anchors, self.fpn_strides):
+        for feat, scale_reg, anchor_centers, stride in zip(
+                feats, self.scales_regs, anchor_centers_split,
+                self.fpn_strides):
             b, _, h, w = get_static_shape(feat)
             inter_feats = []
             for inter_conv in self.inter_convs:
@@ -250,8 +254,8 @@ class TOODHead(nn.Layer):
             # reg prediction and alignment
             reg_dist = scale_reg(self.tood_reg(reg_feat).exp())
             reg_dist = reg_dist.flatten(2).transpose([0, 2, 1])
-            anchor_centers = bbox_center(anchor).unsqueeze(0) / stride
-            reg_bbox = batch_distance2bbox(anchor_centers, reg_dist)
+            reg_bbox = batch_distance2bbox(
+                anchor_centers.unsqueeze(0), reg_dist)
             if self.use_align_head:
                 reg_offset = F.relu(self.reg_offset_conv1(feat))
                 reg_offset = self.reg_offset_conv2(reg_offset)
@@ -268,12 +272,8 @@ class TOODHead(nn.Layer):
             bbox_pred_list.append(bbox_pred)
         cls_score_list = paddle.concat(cls_score_list, axis=1)
         bbox_pred_list = paddle.concat(bbox_pred_list, axis=1)
-        anchors = paddle.concat(anchors)
-        anchors.stop_gradient = True
-        stride_tensor_list = paddle.concat(stride_tensor_list).unsqueeze(0)
-        stride_tensor_list.stop_gradient = True
 
-        return cls_score_list, bbox_pred_list, anchors, num_anchors_list, stride_tensor_list
+        return cls_score_list, bbox_pred_list, anchors, num_anchors_list, stride_tensor
 
     @staticmethod
     def _focal_loss(score, label, alpha=0.25, gamma=2.0):
@@ -287,7 +287,7 @@ class TOODHead(nn.Layer):
 
     def get_loss(self, head_outs, gt_meta):
         pred_scores, pred_bboxes, anchors, \
-        num_anchors_list, stride_tensor_list = head_outs
+        num_anchors_list, stride_tensor = head_outs
         gt_labels = gt_meta['gt_class']
         gt_bboxes = gt_meta['gt_bbox']
         pad_gt_mask = gt_meta['pad_gt_mask']
@@ -304,7 +304,7 @@ class TOODHead(nn.Layer):
         else:
             assigned_labels, assigned_bboxes, assigned_scores = self.assigner(
                 pred_scores.detach(),
-                pred_bboxes.detach() * stride_tensor_list,
+                pred_bboxes.detach() * stride_tensor,
                 bbox_center(anchors),
                 num_anchors_list,
                 gt_labels,
@@ -314,7 +314,7 @@ class TOODHead(nn.Layer):
             alpha_l = -1
 
         # rescale bbox
-        assigned_bboxes /= stride_tensor_list
+        assigned_bboxes /= stride_tensor
         # classification loss
         loss_cls = self._focal_loss(pred_scores, assigned_scores, alpha=alpha_l)
         # select positive samples mask

ppdet/modeling/layers.py

@@ -251,7 +251,7 @@ class LiteConv(nn.Layer):
 
 
 class DropBlock(nn.Layer):
-    def __init__(self, block_size, keep_prob, name, data_format='NCHW'):
+    def __init__(self, block_size, keep_prob, name=None, data_format='NCHW'):
         """
         DropBlock layer, see https://arxiv.org/abs/1810.12890
 

ppdet/modeling/necks/__init__.py

@@ -20,6 +20,7 @@ from . import centernet_fpn
 from . import bifpn
 from . import csp_pan
 from . import es_pan
+from . import custom_pan
 
 from .fpn import *
 from .yolo_fpn import *
@@ -30,3 +31,4 @@ from .blazeface_fpn import *
 from .bifpn import *
 from .csp_pan import *
 from .es_pan import *
+from .custom_pan import *

ppdet/modeling/necks/custom_pan.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from ppdet.core.workspace import register, serializable
+from ppdet.modeling.layers import DropBlock
+from ppdet.modeling.ops import get_act_fn
+from ..backbones.cspresnet import ConvBNLayer, BasicBlock
+from ..shape_spec import ShapeSpec
+
+__all__ = ['CustomCSPPAN']
+
+
+class SPP(nn.Layer):
+    def __init__(self,
+                 ch_in,
+                 ch_out,
+                 k,
+                 pool_size,
+                 act='swish',
+                 data_format='NCHW'):
+        super(SPP, self).__init__()
+        self.pool = []
+        self.data_format = data_format
+        for i, size in enumerate(pool_size):
+            pool = self.add_sublayer(
+                'pool{}'.format(i),
+                nn.MaxPool2D(
+                    kernel_size=size,
+                    stride=1,
+                    padding=size // 2,
+                    data_format=data_format,
+                    ceil_mode=False))
+            self.pool.append(pool)
+        self.conv = ConvBNLayer(ch_in, ch_out, k, padding=k // 2, act=act)
+
+    def forward(self, x):
+        outs = [x]
+        for pool in self.pool:
+            outs.append(pool(x))
+        if self.data_format == 'NCHW':
+            y = paddle.concat(outs, axis=1)
+        else:
+            y = paddle.concat(outs, axis=-1)
+
+        y = self.conv(y)
+        return y
+
+
+class CSPStage(nn.Layer):
+    def __init__(self, block_fn, ch_in, ch_out, n, act='swish', spp=False):
+        super(CSPStage, self).__init__()
+
+        ch_mid = int(ch_out // 2)
+        self.conv1 = ConvBNLayer(ch_in, ch_mid, 1, act=act)
+        self.conv2 = ConvBNLayer(ch_in, ch_mid, 1, act=act)
+        self.convs = nn.Sequential()
+        next_ch_in = ch_mid
+        for i in range(n):
+            self.convs.add_sublayer(
+                str(i),
+                eval(block_fn)(next_ch_in, ch_mid, act=act, shortcut=False))
+            if i == (n - 1) // 2 and spp:
+                self.convs.add_sublayer(
+                    'spp', SPP(ch_mid * 4, ch_mid, 1, [5, 9, 13], act=act))
+            next_ch_in = ch_mid
+        self.conv3 = ConvBNLayer(ch_mid * 2, ch_out, 1, act=act)
+
+    def forward(self, x):
+        y1 = self.conv1(x)
+        y2 = self.conv2(x)
+        y2 = self.convs(y2)
+        y = paddle.concat([y1, y2], axis=1)
+        y = self.conv3(y)
+        return y
+
+
+@register
+@serializable
+class CustomCSPPAN(nn.Layer):
+    __shared__ = ['norm_type', 'data_format', 'width_mult', 'depth_mult', 'trt']
+
+    def __init__(self,
+                 in_channels=[256, 512, 1024],
+                 out_channels=[1024, 512, 256],
+                 norm_type='bn',
+                 act='leaky',
+                 stage_fn='CSPStage',
+                 block_fn='BasicBlock',
+                 stage_num=1,
+                 block_num=3,
+                 drop_block=False,
+                 block_size=3,
+                 keep_prob=0.9,
+                 spp=False,
+                 data_format='NCHW',
+                 width_mult=1.0,
+                 depth_mult=1.0,
+                 trt=False):
+
+        super(CustomCSPPAN, self).__init__()
+        out_channels = [max(round(c * width_mult), 1) for c in out_channels]
+        block_num = max(round(block_num * depth_mult), 1)
+        act = get_act_fn(
+            act, trt=trt) if act is None or isinstance(act,
+                                                       (str, dict)) else act
+        self.num_blocks = len(in_channels)
+        self.data_format = data_format
+        self._out_channels = out_channels
+        in_channels = in_channels[::-1]
+        fpn_stages = []
+        fpn_routes = []
+        for i, (ch_in, ch_out) in enumerate(zip(in_channels, out_channels)):
+            if i > 0:
+                ch_in += ch_pre // 2
+
+            stage = nn.Sequential()
+            for j in range(stage_num):
+                stage.add_sublayer(
+                    str(j),
+                    eval(stage_fn)(block_fn,
+                                   ch_in if j == 0 else ch_out,
+                                   ch_out,
+                                   block_num,
+                                   act=act,
+                                   spp=(spp and i == 0)))
+
+            if drop_block:
+                stage.add_sublayer('drop', DropBlock(block_size, keep_prob))
+
+            fpn_stages.append(stage)
+
+            if i < self.num_blocks - 1:
+                fpn_routes.append(
+                    ConvBNLayer(
+                        ch_in=ch_out,
+                        ch_out=ch_out // 2,
+                        filter_size=1,
+                        stride=1,
+                        padding=0,
+                        act=act))
+
+            ch_pre = ch_out
+
+        self.fpn_stages = nn.LayerList(fpn_stages)
+        self.fpn_routes = nn.LayerList(fpn_routes)
+
+        pan_stages = []
+        pan_routes = []
+        for i in reversed(range(self.num_blocks - 1)):
+            pan_routes.append(
+                ConvBNLayer(
+                    ch_in=out_channels[i + 1],
+                    ch_out=out_channels[i + 1],
+                    filter_size=3,
+                    stride=2,
+                    padding=1,
+                    act=act))
+
+            ch_in = out_channels[i] + out_channels[i + 1]
+            ch_out = out_channels[i]
+            stage = nn.Sequential()
+            for j in range(stage_num):
+                stage.add_sublayer(
+                    str(j),
+                    eval(stage_fn)(block_fn,
+                                   ch_in if j == 0 else ch_out,
+                                   ch_out,
+                                   block_num,
+                                   act=act,
+                                   spp=False))
+            if drop_block:
+                stage.add_sublayer('drop', DropBlock(block_size, keep_prob))
+
+            pan_stages.append(stage)
+
+        self.pan_stages = nn.LayerList(pan_stages[::-1])
+        self.pan_routes = nn.LayerList(pan_routes[::-1])
+
+    def forward(self, blocks, for_mot=False):
+        blocks = blocks[::-1]
+        fpn_feats = []
+
+        for i, block in enumerate(blocks):
+            if i > 0:
+                block = paddle.concat([route, block], axis=1)
+            route = self.fpn_stages[i](block)
+            fpn_feats.append(route)
+
+            if i < self.num_blocks - 1:
+                route = self.fpn_routes[i](route)
+                route = F.interpolate(
+                    route, scale_factor=2., data_format=self.data_format)
+
+        pan_feats = [fpn_feats[-1], ]
+        route = fpn_feats[-1]
+        for i in reversed(range(self.num_blocks - 1)):
+            block = fpn_feats[i]
+            route = self.pan_routes[i](route)
+            block = paddle.concat([route, block], axis=1)
+            route = self.pan_stages[i](block)
+            pan_feats.append(route)
+
+        return pan_feats[::-1]
+
+    @classmethod
+    def from_config(cls, cfg, input_shape):
+        return {'in_channels': [i.channels for i in input_shape], }
+
+    @property
+    def out_shape(self):
+        return [ShapeSpec(channels=c) for c in self._out_channels]

ppdet/modeling/necks/yolo_fpn.py

@@ -114,7 +114,7 @@ class SPP(nn.Layer):
                  ch_out,
                  k,
                  pool_size,
-                 norm_type,
+                 norm_type='bn',
                  freeze_norm=False,
                  name='',
                  act='leaky',

ppdet/modeling/ops.py

@@ -20,28 +20,57 @@ from paddle.regularizer import L2Decay
 
 from paddle.fluid.framework import Variable, in_dygraph_mode
 from paddle.fluid import core
+from paddle.fluid.dygraph import parallel_helper
 from paddle.fluid.layer_helper import LayerHelper
 from paddle.fluid.data_feeder import check_variable_and_dtype, check_type, check_dtype
 
 __all__ = [
-    'roi_pool',
-    'roi_align',
-    'prior_box',
-    'generate_proposals',
-    'iou_similarity',
-    'box_coder',
-    'yolo_box',
-    'multiclass_nms',
-    'distribute_fpn_proposals',
-    'collect_fpn_proposals',
-    'matrix_nms',
-    'batch_norm',
-    'mish',
+    'roi_pool', 'roi_align', 'prior_box', 'generate_proposals',
+    'iou_similarity', 'box_coder', 'yolo_box', 'multiclass_nms',
+    'distribute_fpn_proposals', 'collect_fpn_proposals', 'matrix_nms',
+    'batch_norm', 'mish', 'swish', 'identity'
 ]
 
 
+def identity(x):
+    return x
+
+
 def mish(x):
-    return x * paddle.tanh(F.softplus(x))
+    return F.mish(x) if hasattr(F, mish) else x * F.tanh(F.softplus(x))
+
+
+def swish(x):
+    return x * F.sigmoid(x)
+
+
+TRT_ACT_SPEC = {'swish': swish}
+
+ACT_SPEC = {'mish': mish}
+
+
+def get_act_fn(act=None, trt=False):
+    assert act is None or isinstance(act, (
+        str, dict)), 'name of activation should be str, dict or None'
+    if not act:
+        return identity
+
+    if isinstance(act, dict):
+        name = act['name']
+        act.pop('name')
+        kwargs = act
+    else:
+        name = act
+        kwargs = dict()
+
+    if trt and name in TRT_ACT_SPEC:
+        fn = TRT_ACT_SPEC[name]
+    elif name in ACT_SPEC:
+        fn = ACT_SPEC[name]
+    else:
+        fn = getattr(F, name)
+
+    return lambda x: fn(x, **kwargs)
 
 
 def batch_norm(ch,
@@ -1602,3 +1631,8 @@ def get_static_shape(tensor):
     shape = paddle.shape(tensor)
     shape.stop_gradient = True
     return shape
+
+
+def paddle_distributed_is_initialized():
+    return core.is_compiled_with_dist(
+    ) and parallel_helper._is_parallel_ctx_initialized()

ppdet/optimizer.py

@@ -43,7 +43,7 @@ class CosineDecay(object):
             the max_iters is much larger than the warmup iter
     """
 
-    def __init__(self, max_epochs=1000, use_warmup=True, eta_min=0):
+    def __init__(self, max_epochs=1000, use_warmup=True, eta_min=0.):
         self.max_epochs = max_epochs
         self.use_warmup = use_warmup
         self.eta_min = eta_min
@@ -65,6 +65,7 @@ class CosineDecay(object):
                 decayed_lr = base_lr * 0.5 * (math.cos(
                     (i - warmup_iters) * math.pi /
                     (max_iters - warmup_iters)) + 1)
+                decayed_lr = decayed_lr if decayed_lr > self.eta_min else self.eta_min
                 value.append(decayed_lr)
             return optimizer.lr.PiecewiseDecay(boundary, value)
 
@@ -110,7 +111,7 @@ class PiecewiseDecay(object):
             boundary = [int(step_per_epoch) * i for i in self.milestones]
             value = [base_lr]  # during step[0, boundary[0]] is base_lr
 
-        # self.values is setted directly in config 
+        # self.values is setted directly in config
         if self.values is not None:
             assert len(self.milestones) + 1 == len(self.values)
             return optimizer.lr.PiecewiseDecay(boundary, self.values)
@@ -201,7 +202,7 @@ class LearningRate(object):
             return self.schedulers[0](base_lr=self.base_lr,
                                       step_per_epoch=step_per_epoch)
 
-        # TODO: split warmup & decay 
+        # TODO: split warmup & decay
         # warmup
         boundary, value = self.schedulers[1](self.base_lr, step_per_epoch)
         # decay
@@ -299,10 +300,10 @@ class ModelEMA(object):
             Ema's parameter are updated with the formula:
            `ema_param = decay * ema_param + (1 - decay) * cur_param`.
             Defaults is 0.9998.
-        use_thres_step (bool): Whether set decay by thres_step or not 
-        cycle_epoch (int): The epoch of interval to reset ema_param and 
+        use_thres_step (bool): Whether set decay by thres_step or not
+        cycle_epoch (int): The epoch of interval to reset ema_param and
             step. Defaults is -1, which means not reset. Its function is to
-            add a regular effect to ema, which is set according to experience 
+            add a regular effect to ema, which is set according to experience
             and is effective when the total training epoch is large.
     """
 
@@ -331,6 +332,11 @@ class ModelEMA(object):
         for k, v in self.state_dict.items():
             self.state_dict[k] = paddle.zeros_like(v)
 
+    def resume(self, state_dict, step):
+        for k, v in state_dict.items():
+            self.state_dict[k] = v
+        self.step = step
+
     def update(self, model=None):
         if self.use_thres_step:
             decay = min(self.decay, (1 + self.step) / (10 + self.step))

ppdet/utils/checkpoint.py

@@ -62,7 +62,7 @@ def _strip_postfix(path):
     return path
 
 
-def load_weight(model, weight, optimizer=None):
+def load_weight(model, weight, optimizer=None, ema=None):
     if is_url(weight):
         weight = get_weights_path(weight)
 
@@ -102,6 +102,10 @@ def load_weight(model, weight, optimizer=None):
             last_epoch = optim_state_dict.pop('last_epoch')
         optimizer.set_state_dict(optim_state_dict)
 
+        if ema is not None and os.path.exists(path + '_ema.pdparams'):
+            ema_state_dict = paddle.load(path + '_ema.pdparams')
+            ema.resume(ema_state_dict,
+                       optim_state_dict['LR_Scheduler']['last_epoch'])
     return last_epoch
 
 
@@ -112,9 +116,9 @@ def match_state_dict(model_state_dict, weight_state_dict):
 
     The method supposes that all the names in pretrained weight state dict are
     subclass of the names in models`, if the prefix 'backbone.' in pretrained weight
-    keys is stripped. And we could get the candidates for each model key. Then we 
+    keys is stripped. And we could get the candidates for each model key. Then we
     select the name with the longest matched size as the final match result. For
-    example, the model state dict has the name of 
+    example, the model state dict has the name of
     'backbone.res2.res2a.branch2a.conv.weight' and the pretrained weight as
     name of 'res2.res2a.branch2a.conv.weight' and 'branch2a.conv.weight'. We
     match the 'res2.res2a.branch2a.conv.weight' to the model key.
@@ -125,7 +129,7 @@ def match_state_dict(model_state_dict, weight_state_dict):
 
     def match(a, b):
         if b.startswith('backbone.res5'):
-            # In Faster RCNN, res5 pretrained weights have prefix of backbone, 
+            # In Faster RCNN, res5 pretrained weights have prefix of backbone,
             # however, the corresponding model weights have difficult prefix,
             # bbox_head.
             b = b[9:]
@@ -201,7 +205,7 @@ def load_pretrain_weight(model, pretrain_weight):
     logger.info('Finish loading model weights: {}'.format(weights_path))
 
 
-def save_model(model, optimizer, save_dir, save_name, last_epoch):
+def save_model(model, save_dir, save_name, last_epoch, optimizer=None):
     """
     save model into disk.
 
@@ -224,7 +228,8 @@ def save_model(model, optimizer, save_dir, save_name, last_epoch):
         assert isinstance(model,
                           dict), 'model is not a instance of nn.layer or dict'
         paddle.save(model, save_path + ".pdparams")
-    state_dict = optimizer.state_dict()
-    state_dict['last_epoch'] = last_epoch
-    paddle.save(state_dict, save_path + ".pdopt")
+    if optimizer is not None:
+        state_dict = optimizer.state_dict()
+        state_dict['last_epoch'] = last_epoch
+        paddle.save(state_dict, save_path + ".pdopt")
     logger.info("Save checkpoint: {}".format(save_dir))