# 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. # The code is based on: # https://github.com/open-mmlab/mmdetection/blob/master/mmdet/models/dense_heads/yolox_head.py from __future__ import absolute_import from __future__ import division from __future__ import print_function import math from functools import partial import numpy as np import paddle import paddle.nn as nn import paddle.nn.functional as F from paddle import ParamAttr from paddle.nn.initializer import Normal, Constant from ppdet.core.workspace import register from ppdet.modeling.bbox_utils import distance2bbox, bbox2distance from ppdet.data.transform.atss_assigner import bbox_overlaps from .gfl_head import GFLHead @register class OTAHead(GFLHead): """ OTAHead Args: conv_feat (object): Instance of 'FCOSFeat' num_classes (int): Number of classes fpn_stride (list): The stride of each FPN Layer prior_prob (float): Used to set the bias init for the class prediction layer loss_qfl (object): Instance of QualityFocalLoss. loss_dfl (object): Instance of DistributionFocalLoss. loss_bbox (object): Instance of bbox loss. assigner (object): Instance of label assigner. reg_max: Max value of integral set :math: `{0, ..., reg_max}` n QFL setting. Default: 16. """ __inject__ = [ 'conv_feat', 'dgqp_module', 'loss_class', 'loss_dfl', 'loss_bbox', 'assigner', 'nms' ] __shared__ = ['num_classes'] def __init__(self, conv_feat='FCOSFeat', dgqp_module=None, num_classes=80, fpn_stride=[8, 16, 32, 64, 128], prior_prob=0.01, loss_class='QualityFocalLoss', loss_dfl='DistributionFocalLoss', loss_bbox='GIoULoss', assigner='SimOTAAssigner', reg_max=16, feat_in_chan=256, nms=None, nms_pre=1000, cell_offset=0): super(OTAHead, self).__init__( conv_feat=conv_feat, dgqp_module=dgqp_module, num_classes=num_classes, fpn_stride=fpn_stride, prior_prob=prior_prob, loss_class=loss_class, loss_dfl=loss_dfl, loss_bbox=loss_bbox, reg_max=reg_max, feat_in_chan=feat_in_chan, nms=nms, nms_pre=nms_pre, cell_offset=cell_offset) self.conv_feat = conv_feat self.dgqp_module = dgqp_module self.num_classes = num_classes self.fpn_stride = fpn_stride self.prior_prob = prior_prob self.loss_qfl = loss_class self.loss_dfl = loss_dfl self.loss_bbox = loss_bbox self.reg_max = reg_max self.feat_in_chan = feat_in_chan self.nms = nms self.nms_pre = nms_pre self.cell_offset = cell_offset self.use_sigmoid = self.loss_qfl.use_sigmoid self.assigner = assigner def _get_target_single(self, flatten_cls_pred, flatten_center_and_stride, flatten_bbox, gt_bboxes, gt_labels): """Compute targets for priors in a single image. """ pos_num, label, label_weight, bbox_target = self.assigner( F.sigmoid(flatten_cls_pred), flatten_center_and_stride, flatten_bbox, gt_bboxes, gt_labels) return (pos_num, label, label_weight, bbox_target) def get_loss(self, head_outs, gt_meta): cls_scores, bbox_preds = head_outs num_level_anchors = [ featmap.shape[-2] * featmap.shape[-1] for featmap in cls_scores ] num_imgs = gt_meta['im_id'].shape[0] featmap_sizes = [[featmap.shape[-2], featmap.shape[-1]] for featmap in cls_scores] decode_bbox_preds = [] center_and_strides = [] for featmap_size, stride, bbox_pred in zip(featmap_sizes, self.fpn_stride, bbox_preds): # center in origin image yy, xx = self.get_single_level_center_point(featmap_size, stride, self.cell_offset) center_and_stride = paddle.stack([xx, yy, stride, stride], -1).tile( [num_imgs, 1, 1]) center_and_strides.append(center_and_stride) center_in_feature = center_and_stride.reshape( [-1, 4])[:, :-2] / stride bbox_pred = bbox_pred.transpose([0, 2, 3, 1]).reshape( [num_imgs, -1, 4 * (self.reg_max + 1)]) pred_distances = self.distribution_project(bbox_pred) decode_bbox_pred_wo_stride = distance2bbox( center_in_feature, pred_distances).reshape([num_imgs, -1, 4]) decode_bbox_preds.append(decode_bbox_pred_wo_stride * stride) flatten_cls_preds = [ cls_pred.transpose([0, 2, 3, 1]).reshape( [num_imgs, -1, self.cls_out_channels]) for cls_pred in cls_scores ] flatten_cls_preds = paddle.concat(flatten_cls_preds, axis=1) flatten_bboxes = paddle.concat(decode_bbox_preds, axis=1) flatten_center_and_strides = paddle.concat(center_and_strides, axis=1) gt_boxes, gt_labels = gt_meta['gt_bbox'], gt_meta['gt_class'] pos_num_l, label_l, label_weight_l, bbox_target_l = [], [], [], [] for flatten_cls_pred,flatten_center_and_stride,flatten_bbox,gt_box, gt_label \ in zip(flatten_cls_preds.detach(),flatten_center_and_strides.detach(), \ flatten_bboxes.detach(),gt_boxes, gt_labels): pos_num, label, label_weight, bbox_target = self._get_target_single( flatten_cls_pred, flatten_center_and_stride, flatten_bbox, gt_box, gt_label) pos_num_l.append(pos_num) label_l.append(label) label_weight_l.append(label_weight) bbox_target_l.append(bbox_target) labels = paddle.to_tensor(np.stack(label_l, axis=0)) label_weights = paddle.to_tensor(np.stack(label_weight_l, axis=0)) bbox_targets = paddle.to_tensor(np.stack(bbox_target_l, axis=0)) center_and_strides_list = self._images_to_levels( flatten_center_and_strides, num_level_anchors) labels_list = self._images_to_levels(labels, num_level_anchors) label_weights_list = self._images_to_levels(label_weights, num_level_anchors) bbox_targets_list = self._images_to_levels(bbox_targets, num_level_anchors) num_total_pos = sum(pos_num_l) try: paddle.distributed.all_reduce(num_total_pos) num_total_pos = paddle.clip( num_total_pos / paddle.distributed.get_world_size(), min=1.) except: num_total_pos = max(num_total_pos, 1) loss_bbox_list, loss_dfl_list, loss_qfl_list, avg_factor = [], [], [], [] for cls_score, bbox_pred, center_and_strides, labels, label_weights, bbox_targets, stride in zip( cls_scores, bbox_preds, center_and_strides_list, labels_list, label_weights_list, bbox_targets_list, self.fpn_stride): center_and_strides = center_and_strides.reshape([-1, 4]) cls_score = cls_score.transpose([0, 2, 3, 1]).reshape( [-1, self.cls_out_channels]) bbox_pred = bbox_pred.transpose([0, 2, 3, 1]).reshape( [-1, 4 * (self.reg_max + 1)]) bbox_targets = bbox_targets.reshape([-1, 4]) labels = labels.reshape([-1]) label_weights = label_weights.reshape([-1]) bg_class_ind = self.num_classes pos_inds = paddle.nonzero( paddle.logical_and((labels >= 0), (labels < bg_class_ind)), as_tuple=False).squeeze(1) score = np.zeros(labels.shape) if len(pos_inds) > 0: pos_bbox_targets = paddle.gather(bbox_targets, pos_inds, axis=0) pos_bbox_pred = paddle.gather(bbox_pred, pos_inds, axis=0) pos_centers = paddle.gather( center_and_strides[:, :-2], pos_inds, axis=0) / stride weight_targets = F.sigmoid(cls_score.detach()) weight_targets = paddle.gather( weight_targets.max(axis=1, keepdim=True), pos_inds, axis=0) pos_bbox_pred_corners = self.distribution_project(pos_bbox_pred) pos_decode_bbox_pred = distance2bbox(pos_centers, pos_bbox_pred_corners) pos_decode_bbox_targets = pos_bbox_targets / stride bbox_iou = bbox_overlaps( pos_decode_bbox_pred.detach().numpy(), pos_decode_bbox_targets.detach().numpy(), is_aligned=True) score[pos_inds.numpy()] = bbox_iou pred_corners = pos_bbox_pred.reshape([-1, self.reg_max + 1]) target_corners = bbox2distance(pos_centers, pos_decode_bbox_targets, self.reg_max).reshape([-1]) # regression loss loss_bbox = paddle.sum( self.loss_bbox(pos_decode_bbox_pred, pos_decode_bbox_targets) * weight_targets) # dfl loss loss_dfl = self.loss_dfl( pred_corners, target_corners, weight=weight_targets.expand([-1, 4]).reshape([-1]), avg_factor=4.0) else: loss_bbox = bbox_pred.sum() * 0 loss_dfl = bbox_pred.sum() * 0 weight_targets = paddle.to_tensor([0], dtype='float32') # qfl loss score = paddle.to_tensor(score) loss_qfl = self.loss_qfl( cls_score, (labels, score), weight=label_weights, avg_factor=num_total_pos) loss_bbox_list.append(loss_bbox) loss_dfl_list.append(loss_dfl) loss_qfl_list.append(loss_qfl) avg_factor.append(weight_targets.sum()) avg_factor = sum(avg_factor) try: paddle.distributed.all_reduce(avg_factor) avg_factor = paddle.clip( avg_factor / paddle.distributed.get_world_size(), min=1) except: avg_factor = max(avg_factor.item(), 1) if avg_factor <= 0: loss_qfl = paddle.to_tensor(0, dtype='float32', stop_gradient=False) loss_bbox = paddle.to_tensor( 0, dtype='float32', stop_gradient=False) loss_dfl = paddle.to_tensor(0, dtype='float32', stop_gradient=False) else: losses_bbox = list(map(lambda x: x / avg_factor, loss_bbox_list)) losses_dfl = list(map(lambda x: x / avg_factor, loss_dfl_list)) loss_qfl = sum(loss_qfl_list) loss_bbox = sum(losses_bbox) loss_dfl = sum(losses_dfl) loss_states = dict( loss_qfl=loss_qfl, loss_bbox=loss_bbox, loss_dfl=loss_dfl) return loss_states @register class OTAVFLHead(OTAHead): __inject__ = [ 'conv_feat', 'dgqp_module', 'loss_class', 'loss_dfl', 'loss_bbox', 'assigner', 'nms' ] __shared__ = ['num_classes'] def __init__(self, conv_feat='FCOSFeat', dgqp_module=None, num_classes=80, fpn_stride=[8, 16, 32, 64, 128], prior_prob=0.01, loss_class='VarifocalLoss', loss_dfl='DistributionFocalLoss', loss_bbox='GIoULoss', assigner='SimOTAAssigner', reg_max=16, feat_in_chan=256, nms=None, nms_pre=1000, cell_offset=0): super(OTAVFLHead, self).__init__( conv_feat=conv_feat, dgqp_module=dgqp_module, num_classes=num_classes, fpn_stride=fpn_stride, prior_prob=prior_prob, loss_class=loss_class, loss_dfl=loss_dfl, loss_bbox=loss_bbox, reg_max=reg_max, feat_in_chan=feat_in_chan, nms=nms, nms_pre=nms_pre, cell_offset=cell_offset) self.conv_feat = conv_feat self.dgqp_module = dgqp_module self.num_classes = num_classes self.fpn_stride = fpn_stride self.prior_prob = prior_prob self.loss_vfl = loss_class self.loss_dfl = loss_dfl self.loss_bbox = loss_bbox self.reg_max = reg_max self.feat_in_chan = feat_in_chan self.nms = nms self.nms_pre = nms_pre self.cell_offset = cell_offset self.use_sigmoid = self.loss_vfl.use_sigmoid self.assigner = assigner def get_loss(self, head_outs, gt_meta): cls_scores, bbox_preds = head_outs num_level_anchors = [ featmap.shape[-2] * featmap.shape[-1] for featmap in cls_scores ] num_imgs = gt_meta['im_id'].shape[0] featmap_sizes = [[featmap.shape[-2], featmap.shape[-1]] for featmap in cls_scores] decode_bbox_preds = [] center_and_strides = [] for featmap_size, stride, bbox_pred in zip(featmap_sizes, self.fpn_stride, bbox_preds): # center in origin image yy, xx = self.get_single_level_center_point(featmap_size, stride, self.cell_offset) strides = paddle.full((len(xx), ), stride) center_and_stride = paddle.stack([xx, yy, strides, strides], -1).tile([num_imgs, 1, 1]) center_and_strides.append(center_and_stride) center_in_feature = center_and_stride.reshape( [-1, 4])[:, :-2] / stride bbox_pred = bbox_pred.transpose([0, 2, 3, 1]).reshape( [num_imgs, -1, 4 * (self.reg_max + 1)]) pred_distances = self.distribution_project(bbox_pred) decode_bbox_pred_wo_stride = distance2bbox( center_in_feature, pred_distances).reshape([num_imgs, -1, 4]) decode_bbox_preds.append(decode_bbox_pred_wo_stride * stride) flatten_cls_preds = [ cls_pred.transpose([0, 2, 3, 1]).reshape( [num_imgs, -1, self.cls_out_channels]) for cls_pred in cls_scores ] flatten_cls_preds = paddle.concat(flatten_cls_preds, axis=1) flatten_bboxes = paddle.concat(decode_bbox_preds, axis=1) flatten_center_and_strides = paddle.concat(center_and_strides, axis=1) gt_boxes, gt_labels = gt_meta['gt_bbox'], gt_meta['gt_class'] pos_num_l, label_l, label_weight_l, bbox_target_l = [], [], [], [] for flatten_cls_pred, flatten_center_and_stride, flatten_bbox,gt_box,gt_label \ in zip(flatten_cls_preds.detach(), flatten_center_and_strides.detach(), \ flatten_bboxes.detach(),gt_boxes,gt_labels): pos_num, label, label_weight, bbox_target = self._get_target_single( flatten_cls_pred, flatten_center_and_stride, flatten_bbox, gt_box, gt_label) pos_num_l.append(pos_num) label_l.append(label) label_weight_l.append(label_weight) bbox_target_l.append(bbox_target) labels = paddle.to_tensor(np.stack(label_l, axis=0)) label_weights = paddle.to_tensor(np.stack(label_weight_l, axis=0)) bbox_targets = paddle.to_tensor(np.stack(bbox_target_l, axis=0)) center_and_strides_list = self._images_to_levels( flatten_center_and_strides, num_level_anchors) labels_list = self._images_to_levels(labels, num_level_anchors) label_weights_list = self._images_to_levels(label_weights, num_level_anchors) bbox_targets_list = self._images_to_levels(bbox_targets, num_level_anchors) num_total_pos = sum(pos_num_l) try: paddle.distributed.all_reduce(num_total_pos) num_total_pos = paddle.clip( num_total_pos / paddle.distributed.get_world_size(), min=1.) except: num_total_pos = max(num_total_pos, 1) loss_bbox_list, loss_dfl_list, loss_vfl_list, avg_factor = [], [], [], [] for cls_score, bbox_pred, center_and_strides, labels, label_weights, bbox_targets, stride in zip( cls_scores, bbox_preds, center_and_strides_list, labels_list, label_weights_list, bbox_targets_list, self.fpn_stride): center_and_strides = center_and_strides.reshape([-1, 4]) cls_score = cls_score.transpose([0, 2, 3, 1]).reshape( [-1, self.cls_out_channels]) bbox_pred = bbox_pred.transpose([0, 2, 3, 1]).reshape( [-1, 4 * (self.reg_max + 1)]) bbox_targets = bbox_targets.reshape([-1, 4]) labels = labels.reshape([-1]) bg_class_ind = self.num_classes pos_inds = paddle.nonzero( paddle.logical_and((labels >= 0), (labels < bg_class_ind)), as_tuple=False).squeeze(1) # vfl vfl_score = np.zeros(cls_score.shape) if len(pos_inds) > 0: pos_bbox_targets = paddle.gather(bbox_targets, pos_inds, axis=0) pos_bbox_pred = paddle.gather(bbox_pred, pos_inds, axis=0) pos_centers = paddle.gather( center_and_strides[:, :-2], pos_inds, axis=0) / stride weight_targets = F.sigmoid(cls_score.detach()) weight_targets = paddle.gather( weight_targets.max(axis=1, keepdim=True), pos_inds, axis=0) pos_bbox_pred_corners = self.distribution_project(pos_bbox_pred) pos_decode_bbox_pred = distance2bbox(pos_centers, pos_bbox_pred_corners) pos_decode_bbox_targets = pos_bbox_targets / stride bbox_iou = bbox_overlaps( pos_decode_bbox_pred.detach().numpy(), pos_decode_bbox_targets.detach().numpy(), is_aligned=True) # vfl pos_labels = paddle.gather(labels, pos_inds, axis=0) vfl_score[pos_inds.numpy(), pos_labels] = bbox_iou pred_corners = pos_bbox_pred.reshape([-1, self.reg_max + 1]) target_corners = bbox2distance(pos_centers, pos_decode_bbox_targets, self.reg_max).reshape([-1]) # regression loss loss_bbox = paddle.sum( self.loss_bbox(pos_decode_bbox_pred, pos_decode_bbox_targets) * weight_targets) # dfl loss loss_dfl = self.loss_dfl( pred_corners, target_corners, weight=weight_targets.expand([-1, 4]).reshape([-1]), avg_factor=4.0) else: loss_bbox = bbox_pred.sum() * 0 loss_dfl = bbox_pred.sum() * 0 weight_targets = paddle.to_tensor([0], dtype='float32') # vfl loss num_pos_avg_per_gpu = num_total_pos vfl_score = paddle.to_tensor(vfl_score) loss_vfl = self.loss_vfl( cls_score, vfl_score, avg_factor=num_pos_avg_per_gpu) loss_bbox_list.append(loss_bbox) loss_dfl_list.append(loss_dfl) loss_vfl_list.append(loss_vfl) avg_factor.append(weight_targets.sum()) avg_factor = sum(avg_factor) try: paddle.distributed.all_reduce(avg_factor) avg_factor = paddle.clip( avg_factor / paddle.distributed.get_world_size(), min=1) except: avg_factor = max(avg_factor.item(), 1) if avg_factor <= 0: loss_vfl = paddle.to_tensor(0, dtype='float32', stop_gradient=False) loss_bbox = paddle.to_tensor( 0, dtype='float32', stop_gradient=False) loss_dfl = paddle.to_tensor(0, dtype='float32', stop_gradient=False) else: losses_bbox = list(map(lambda x: x / avg_factor, loss_bbox_list)) losses_dfl = list(map(lambda x: x / avg_factor, loss_dfl_list)) loss_vfl = sum(loss_vfl_list) loss_bbox = sum(losses_bbox) loss_dfl = sum(losses_dfl) loss_states = dict( loss_vfl=loss_vfl, loss_bbox=loss_bbox, loss_dfl=loss_dfl) return loss_states