Home » Football » Germany » DFB Junioren Pokal

Estadísticas y predicciones de DFB Junioren Pokal

Today

< div class = " team-history " > < h3 > Bayer Leverkusen < / h3 > < ul > < li > Ganó el DFB-Junioren-Pokal hace cinco temporadas.< / li > < li > Subcampeón de la Bundesliga Juvenil hace dos temporadas.< / li > < li > Tiene una racha positiva con 6 victorias consecutivas.< / li > <|file_sep|># -*- coding: utf-8 -*- import torch.nn as nn from model.blocks import BasicBlock class ResNet(nn.Module): def __init__(self): super(ResNet, self).__init__() self.in_channels = 64 self.conv1 = nn.Sequential( nn.Conv2d(3, self.in_channels, kernel_size=7, stride=1, padding=3), nn.BatchNorm2d(self.in_channels), nn.ReLU(), nn.MaxPool2d(kernel_size=4, stride=4) ) self.conv_layers = self._make_layers() self.fc = nn.Linear(64 * BasicBlock.expansion, 128) def _make_layers(self): layers = [] layers.append(BasicBlock(in_channels=self.in_channels, out_channels=64)) layers.append(BasicBlock(in_channels=64 * BasicBlock.expansion, out_channels=128)) layers.append(BasicBlock(in_channels=128 * BasicBlock.expansion, out_channels=256)) layers.append(BasicBlock(in_channels=256 * BasicBlock.expansion, out_channels=512)) return nn.Sequential(*layers) def forward(self, x): out = self.conv1(x) out = self.conv_layers(out) out = nn.AdaptiveAvgPool2d((1, 1))(out) out = out.view(out.size(0), -1) out = self.fc(out) return out def resnet(): return ResNet() if __name__ == '__main__': net = resnet() print(net) <|repo_name|>XiaoZhouMeng/RPN<|file_sep|>/model/loss.py # -*- coding: utf-8 -*- import torch def rpn_loss(pred_cls_prob, gt_cls_label, pred_bbox_delta, gt_bbox_label): pred_cls_prob = torch.squeeze(pred_cls_prob) gt_cls_label = torch.squeeze(gt_cls_label) pred_bbox_delta = torch.squeeze(pred_bbox_delta) gt_bbox_label = torch.squeeze(gt_bbox_label) pos_num = torch.sum(gt_cls_label == 1).item() neg_num = torch.sum(gt_cls_label == -1).item() batch_size = gt_cls_label.shape[0] pos_mask = (gt_cls_label == 1).float() neg_mask = (gt_cls_label == -1).float() obj_mask = pos_mask + neg_mask pos_loss_weight = batch_size * neg_num / (batch_size * pos_num + 1e-6) # 将正例的loss放大，减少负例的loss，使得模型更加关注正例的损失 bbox_mask = (gt_bbox_label != -1).float() total_loss_pos_weighted_cls = torch.sum( pos_loss_weight * pos_mask * torch.nn.functional.binary_cross_entropy_with_logits( pred_cls_prob[obj_mask == 1], gt_cls_label[obj_mask == 1], reduction='none' )) if pos_num != 0 else torch.tensor(0.).cuda() total_loss_neg_weighted_cls = torch.sum( neg_mask * torch.nn.functional.binary_cross_entropy_with_logits( pred_cls_prob[obj_mask == -1], gt_cls_label[obj_mask == -1], reduction='none' )) if neg_num != 0 else torch.tensor(0.).cuda() total_loss_pos_weighted_regu = torch.sum( bbox_mask * torch.nn.functional.smooth_l1_loss( pred_bbox_delta[obj_mask == 1], gt_bbox_label[obj_mask == 1], reduction='none' )) if pos_num != 0 else torch.tensor(0.).cuda() total_loss_pos_weighted_regu /= (pos_num + 1e-6) total_loss_neg_weighted_regu = torch.tensor(0.).cuda() return total_loss_pos_weighted_cls + total_loss_neg_weighted_cls + total_loss_pos_weighted_regu + total_loss_neg_weighted_regu def fast_rcnn_loss(pred_cls_prob, gt_cls_label, pred_bbox_delta, gt_bbox_label): pred_cls_prob = torch.squeeze(pred_cls_prob) gt_cls_label = torch.squeeze(gt_cls_label) pred_bbox_delta = torch.squeeze(pred_bbox_delta) gt_bbox_label = torch.squeeze(gt_bbox_label) obj_num_per_image_batch_size_list = [] for i in range(len(gt_cls_label)): obj_num_per_image_batch_size_list.append(torch.sum(gt_cls_label[i] >= 0).item()) obj_num_per_image_batch_size_tensor_list = [] for i in range(len(obj_num_per_image_batch_size_list)): obj_num_per_image_batch_size_tensor_list.append(torch.tensor([obj_num_per_image_batch_size_list[i]]).cuda()) total_loss_pos_weighted_regu_list = [] total_loss_pos_weighted_regu_tensor_list = [] for i in range(len(obj_num_per_image_batch_size_tensor_list)): total_loss_pos_weighted_regu_list.append(torch.sum( torch.nn.functional.smooth_l1_loss( pred_bbox_delta[i][gt_cls_label[i] >= 0], gt_bbox_label[i][gt_cls_label[i] >= 0], reduction='none' ))) for i in range(len(total_loss_pos_weighted_regu_list)): total_loss_pos_weighted_regu_tensor_list.append(total_loss_pos_weighted_regu_list[i] / (obj_num_per_image_batch_size_tensor_list[i] + 1e-6)) total_loss_pos_weighted_regu_tensor_stack_list = [] for i in range(len(total_loss_pos_weighted_regu_tensor_list)): total_loss_pos_weighted_regu_tensor_stack_list.append(total_loss_pos_weighted_regu_tensor_list[i].unsqueeze(0)) total_loss_pos_weighted_regu_stack_tensor_stack_list_summed_over_batch_dim_zero_tensor_stack_reshaped_to_original_dim_zero_dim_one_dim_two_dim_three_and_squeezed_onto_dim_zero_reshaped_to_original_dim_one_dim_two_dim_three_and_squeezed_onto_dim_zero_tensor_stack_reshaped_to_original_dim_zero_dim_one_dim_two_dim_three_and_squeezed_onto_dim_zero_reshaped_to_original_dim_one_dim_two_dim_three_and_squeezed_onto_dim_zero_tensor_stack_reshaped_to_original_dim_zero_dim_one_dim_two_dim_three_and_squeezed_onto_dim_zero_reshaped_to_original_dim_one_dim_two_dim_three_and_squeezed_onto_dim_zero_reshaped_to_original__tensor_stack_summed_over_batch_dimension_and_squeezed_dimensionless_tensor_stack_reshaped_to_original__tensor_stack_summed_over_batch_dimension_and_squeezed_dimensionless_tensor_stack_reshaped_to_original__tensor_stack_summed_over_batch_dimension_and_squeezed_dimensionless_tensor_stack_reshaped_to_original__tensor