# 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 from ppdet.modeling.layers import MultiClassNMS __all__ = ['PPYOLOEHead'] 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 PPYOLOEHead(nn.Layer): __shared__ = ['num_classes', 'eval_size', 'trt', 'exclude_nms'] __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_size=None, loss_weight={ 'class': 1.0, 'iou': 2.5, 'dfl': 0.5, }, trt=False, exclude_nms=False): super(PPYOLOEHead, 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_size = eval_size self.static_assigner_epoch = static_assigner_epoch self.static_assigner = static_assigner self.assigner = assigner self.nms = nms if isinstance(self.nms, MultiClassNMS) and trt: self.nms.trt = trt self.exclude_nms = exclude_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.proj_conv.skip_quant = True 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_size: anchor_points, stride_tensor = self._generate_anchors() self.anchor_points = anchor_points self.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_size[0] / stride) w = int(self.eval_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_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 = pred_dist.sum() * 0. 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 + 1)[..., :-1] loss_cls = self._varifocal_loss(pred_scores, assigned_scores, one_hot_label) else: loss_cls = self._focal_loss(pred_scores, assigned_scores, 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 if self.exclude_nms: # `exclude_nms=True` just use in benchmark return pred_bboxes.sum(), pred_scores.sum() else: bbox_pred, bbox_num, _ = self.nms(pred_bboxes, pred_scores) return bbox_pred, bbox_num