Open Malledive opened 3 years ago
loading annotations into memory... Done (t=0.00s) creating index... index created! loading annotations into memory... Done (t=0.00s) creating index... index created! [Info] initializing weights... Step: 36. Epoch: 0/500. Iteration: 37/37. Cls loss: 152.74318. Reg loss: 0.00000. Total loss: 152.74318: 100% 37/37 [00:32<00:00, 1.15it/s] Val. Epoch: 0/500. Classification loss: 0.05725. Regression loss: 0.00000. Total loss: 0.05725 Step: 73. Epoch: 1/500. Iteration: 37/37. Cls loss: 58.51061. Reg loss: 0.00000. Total loss: 58.51061: 100% 37/37 [00:32<00:00, 1.14it/s] Val. Epoch: 1/500. Classification loss: 0.05707. Regression loss: 0.00000. Total loss: 0.05707 Step: 110. Epoch: 2/500. Iteration: 37/37. Cls loss: 44.43995. Reg loss: 0.00000. Total loss: 44.43995: 100% 37/37 [00:32<00:00, 1.14it/s] Val. Epoch: 2/500. Classification loss: 0.05713. Regression loss: 0.00000. Total loss: 0.05713 Step: 147. Epoch: 3/500. Iteration: 37/37. Cls loss: 46.96254. Reg loss: 0.00000. Total loss: 46.96254: 100% 37/37 [00:32<00:00, 1.14it/s] Val. Epoch: 3/500. Classification loss: 0.05734. Regression loss: 0.00000. Total loss: 0.05734 Step: 184. Epoch: 4/500. Iteration: 37/37. Cls loss: 34.39653. Reg loss: 0.00000. Total loss: 34.39653: 100% 37/37 [00:32<00:00, 1.13it/s] Val. Epoch: 4/500. Classification loss: 0.05767. Regression loss: 0.00000. Total loss: 0.05767 Step: 221. Epoch: 5/500. Iteration: 37/37. Cls loss: 18.23889. Reg loss: 0.00000. Total loss: 18.23889: 100% 37/37 [00:32<00:00, 1.14it/s] Val. Epoch: 5/500. Classification loss: 0.05805. Regression loss: 0.00000. Total loss: 0.05805 Step: 258. Epoch: 6/500. Iteration: 37/37. Cls loss: 23.31059. Reg loss: 0.00000. Total loss: 23.31059: 100% 37/37 [00:32<00:00, 1.13it/s] Val. Epoch: 6/500. Classification loss: 0.05858. Regression loss: 0.00000. Total loss: 0.05858 Step: 295. Epoch: 7/500. Iteration: 37/37. Cls loss: 17.64239. Reg loss: 0.00000. Total loss: 17.64239: 100% 37/37 [00:32<00:00, 1.14it/s] Val. Epoch: 7/500. Classification loss: 0.05914. Regression loss: 0.00000. Total loss: 0.05914 Step: 332. Epoch: 8/500. Iteration: 37/37. Cls loss: 13.08778. Reg loss: 0.00000. Total loss: 13.08778: 100% 37/37 [00:32<00:00, 1.14it/s] Val. Epoch: 8/500. Classification loss: 0.05977. Regression loss: 0.00000. Total loss: 0.05977
Building a model from scratch, but the Regression loss stays at zero
It now works with pre-trained models, but:
RuntimeError: Error(s) in loading state_dict for EfficientDetBackbone: Missing key(s) in state_dict: "bifpn.5.p6_w1", "bifpn.5.p5_w1", "bifpn.5.p4_w1", "bifpn.5.p3_w1", "bifpn.5.p4_w2", "bifpn.5.p5_w2", "bifpn.5.p6_w2", "bifpn.5.p7_w2", "bifpn.5.conv6_up.depthwise_conv.conv.weight", "bifpn.5.conv6_up.pointwise_conv.conv.weight", "bifpn.5.conv6_up.pointwise_conv.conv.bias", "bifpn.5.conv6_up.bn.weight", "bifpn.5.conv6_up.bn.bias", "bifpn.5.conv6_up.bn.running_mean", "bifpn.5.conv6_up.bn.running_var", "bifpn.5.conv5_up.depthwise_conv.conv.weight", "bifpn.5.conv5_up.pointwise_conv.conv.weight", "bifpn.5.conv5_up.pointwise_conv.conv.bias", "bifpn.5.conv5_up.bn.weight", "bifpn.5.conv5_up.bn.bias", "bifpn.5.conv5_up.bn.running_mean", "bifpn.5.conv5_up.bn.running_var", "bifpn.5.conv4_up.depthwise_conv.conv.weight", "bifpn.5.conv4_up.pointwise_conv.conv.weight", "bifpn.5.conv4_up.pointwise_conv.conv.bias", "bifpn.5.conv4_up.bn.weight", "bifpn.5.conv4_up.bn.bias", "bifpn.5.conv4_up.bn.running_mean", "bifpn.5.conv4_up.bn.running_var", "bifpn.5.conv3_up.depthwise_conv.conv.weight", "bifpn.5.conv3_up.pointwise_conv.conv.weight", "bifpn.5.conv3_up.pointwise_conv.conv.bias", "bifpn.5.conv3_up.bn.weight", "bifpn.5.conv3_up.bn.bias", "bifpn.5.conv3_up.bn.running_mean", "bifpn.5.conv3_up.bn.running_var", "bifpn.5.conv4_down.depthwise_conv.conv.weight", "bifpn.5.conv4_down.pointwise_conv.conv.weight", "bifpn.5.conv4_down.pointwise_conv.conv.bias", "bifpn.5.conv4_down.bn.weight", "bifpn.5.conv4_down.bn.bias", "bifpn.5.conv4_down.bn.running_mean", "bifpn.5.conv4_down.bn.running_var", "bifpn.5.conv5_down.depthwise_conv.conv.weight", "bifpn.5.conv5_down.pointwise_conv.conv.weight", "bifpn.5.conv5_down.pointwise_conv.conv.bias", "bifpn.5.conv5_down.bn.weight", "bifpn.5.conv5_down.bn.bias", "bifpn.5.conv5_down.bn.running_mean", "bifpn.5.conv5_down.bn.running_var", "bifpn.5.conv6_down.depthwise_conv.conv.weight", "bifpn.5.conv6_down.pointwise_conv.conv.weight", "bifpn.5.conv6_down.pointwise_conv.conv.bias", "bifpn.5.conv6_down.bn.weight", "bifpn.5.conv6_down.bn.bias", "bifpn.5.conv6_down.bn.running_mean", "bifpn.5.conv6_down.bn.running_var", "bifpn.5.conv7_down.depthwise_conv.conv.weight", "bifpn.5.conv7_down.pointwise_conv.conv.weight", "bifpn.5.conv7_down.pointwise_conv.conv.bias", "bifpn.5.conv7_down.bn.weight", "bifpn.5.conv7_down.bn.bias", "bifpn.5.conv7_down.bn.running_mean", "bifpn.5.conv7_down.bn.running_var", "bifpn.6.p6_w1", "bifpn.6.p5_w1", "bifpn.6.p4_w1", "bifpn.6.p3_w1", "bifpn.6.p4_w2", "bifpn.6.p5_w2", "bifpn.6.p6_w2", "bifpn.6.p7_w2", "bifpn.6.conv6_up.depthwise_conv.conv.weight", "bifpn.6.conv6_up.pointwise_conv.conv.weight", "bifpn.6.conv6_up.pointwise_conv.conv.bias", "bifpn.6.conv6_up.bn.weight", "bifpn.6.conv6_up.bn.bias", "bifpn.6.conv6_up.bn.running_mean", "bifpn.6.conv6_up.bn.running_var", "bifpn.6.conv5_up.depthwise_conv.conv.weight", "bifpn.6.conv5_up.pointwise_conv.conv.weight", "bifpn.6.conv5_up.pointwise_conv.conv.bias", "bifpn.6.conv5_up.bn.weight", "bifpn.6.conv5_up.bn.bias", "bifpn.6.conv5_up.bn.running_mean", "bifpn.6.conv5_up.bn.running_var", "bifpn.6.conv4_up.depthwise_conv.conv.weight", "bifpn.6.conv4_up.pointwise_conv.conv.weight", "bifpn.6.conv4_up.pointwise_conv.conv.bias", "bifpn.6.conv4_up.bn.weight", "bifpn.6.conv4_up.bn.bias", "bifpn.6.conv4_up.bn.running_mean", "bifpn.6.conv4_up.bn.running_var", "bifpn.6.conv3_up.depthwise_conv.conv.weight", "bifpn.6.conv3_up.pointwise_conv.conv.weight", "bifpn.6.conv3_up.pointwise_conv.conv.bias", "bifpn.6.conv3_up.bn.weight", "bifpn.6.conv3_up.bn.bias", "bifpn.6.conv3_up.bn.running_mean", "bifpn.6.conv3_up.bn.running_var", "bifpn.6.conv4_down.depthwise_conv.conv.weight", "bifpn.6.conv4_down.pointwise_conv.conv.weight", "bifpn.6.conv4_down.pointwise_conv.conv.bias", "bifpn.6.conv4_down.bn.weight", "bifpn.6.conv4_down.bn.bias", "bifpn.6.conv4_down.bn.running_mean", "bifpn.6.conv4_down.bn.running_var", "bifpn.6.conv5_down.depthwise_conv.conv.weight", "bifpn.6.conv5_down.pointwise_conv.conv.weight", "bifpn.6.conv5_down.pointwise_conv.conv.bias", "bifpn.6.conv5_down.bn.weight", "bifpn.6.conv5_down.bn.bias", "bifpn.6.conv5_down.bn.running_mean", "bifpn.6.conv5_down.bn.running_var", "bifpn.6.conv6_down.depthwise_conv.conv.weight", "bifpn.6.conv6_down.pointwise_conv.conv.weight", "bifpn.6.conv6_down.pointwise_conv.conv.bias", "bifpn.6.conv6_down.bn.weight", "bifpn.6.conv6_down.bn.bias", "bifpn.6.conv6_down.bn.running_mean", "bifpn.6.conv6_down.bn.running_var", "bifpn.6.conv7_down.depthwise_conv.conv.weight", "bifpn.6.conv7_down.pointwise_conv.conv.weight", "bifpn.6.conv7_down.pointwise_conv.conv.bias", "bifpn.6.conv7_down.bn.weight", "bifpn.6.conv7_down.bn.bias", "bifpn.6.conv7_down.bn.running_mean", "bifpn.6.conv7_down.bn.running_var", "regressor.conv_list.3.depthwise_conv.conv.weight", "regressor.conv_list.3.pointwise_conv.conv.weight", "regressor.conv_list.3.pointwise_conv.conv.bias", "regressor.bn_list.0.3.weight", "regressor.bn_list.0.3.bias", "regressor.bn_list.0.3.running_mean", "regressor.bn_list.0.3.running_var", "regressor.bn_list.1.3.weight", "regressor.bn_list.1.3.bias", "regressor.bn_list.1.3.running_mean", "regressor.bn_list.1.3.running_var", "regressor.bn_list.2.3.weight", "regressor.bn_list.2.3.bias", "regressor.bn_list.2.3.running_mean", "regressor.bn_list.2.3.running_var", "regressor.bn_list.3.3.weight", "regressor.bn_list.3.3.bias", "regressor.bn_list.3.3.running_mean", "regressor.bn_list.3.3.running_var", "regressor.bn_list.4.3.weight", "regressor.bn_list.4.3.bias", "regressor.bn_list.4.3.running_mean", "regressor.bn_list.4.3.running_var", "classifier.conv_list.3.depthwise_conv.conv.weight", "classifier.conv_list.3.pointwise_conv.conv.weight", "classifier.conv_list.3.pointwise_conv.conv.bias", "classifier.bn_list.0.3.weight", "classifier.bn_list.0.3.bias", "classifier.bn_list.0.3.running_mean", "classifier.bn_list.0.3.running_var", "classifier.bn_list.1.3.weight", "classifier.bn_list.1.3.bias", "classifier.bn_list.1.3.running_mean", "classifier.bn_list.1.3.running_var", "classifier.bn_list.2.3.weight", "classifier.bn_list.2.3.bias", "classifier.bn_list.2.3.running_mean", "classifier.bn_list.2.3.running_var", "classifier.bn_list.3.3.weight", "classifier.bn_list.3.3.bias", "classifier.bn_list.3.3.running_mean", "classifier.bn_list.3.3.running_var", "classifier.bn_list.4.3.weight", "classifier.bn_list.4.3.bias", "classifier.bn_list.4.3.running_mean", "classifier.bn_list.4.3.running_var", "backbone_net.model._blocks.23._expand_conv.conv.weight", "backbone_net.model._blocks.23._bn0.weight", "backbone_net.model._blocks.23._bn0.bias", "backbone_net.model._blocks.23._bn0.running_mean", "backbone_net.model._blocks.23._bn0.running_var", "backbone_net.model._blocks.23._depthwise_conv.conv.weight", "backbone_net.model._blocks.23._bn1.weight", "backbone_net.model._blocks.23._bn1.bias", "backbone_net.model._blocks.23._bn1.running_mean", "backbone_net.model._blocks.23._bn1.running_var", "backbone_net.model._blocks.23._se_reduce.conv.weight", "backbone_net.model._blocks.23._se_reduce.conv.bias", "backbone_net.model._blocks.23._se_expand.conv.weight", "backbone_net.model._blocks.23._se_expand.conv.bias", "backbone_net.model._blocks.23._project_conv.conv.weight", "backbone_net.model._blocks.23._bn2.weight", "backbone_net.model._blocks.23._bn2.bias", "backbone_net.model._blocks.23._bn2.running_mean", "backbone_net.model._blocks.23._bn2.running_var", "backbone_net.model._blocks.24._expand_conv.conv.weight", "backbone_net.model._blocks.24._bn0.weight", "backbone_net.model._blocks.24._bn0.bias", "backbone_net.model._blocks.24._bn0.running_mean", "backbone_net.model._blocks.24._bn0.running_var", "backbone_net.model._blocks.24._depthwise_conv.conv.weight", "backbone_net.model._blocks.24._bn1.weight", "backbone_net.model._blocks.24._bn1.bias", "backbone_net.model._blocks.24._bn1.running_mean", "backbone_net.model._blocks.24._bn1.running_var", "backbone_net.model._blocks.24._se_reduce.conv.weight", "backbone_net.model._blocks.24._se_reduce.conv.bias", "backbone_net.model._blocks.24._se_expand.conv.weight", "backbone_net.model._blocks.24._se_expand.conv.bias", "backbone_net.model._blocks.24._project_conv.conv.weight", "backbone_net.model._blocks.24._bn2.weight", "backbone_net.model._blocks.24._bn2.bias", "backbone_net.model._blocks.24._bn2.running_mean", "backbone_net.model._blocks.24._bn2.running_var", "backbone_net.model._blocks.25._expand_conv.conv.weight", "backbone_net.model._blocks.25._bn0.weight", "backbone_net.model._blocks.25._bn0.bias", "backbone_net.model._blocks.25._bn0.running_mean", "backbone_net.model._blocks.25._bn0.running_var", "backbone_net.model._blocks.25._depthwise_conv.conv.weight", "backbone_net.model._blocks.25._bn1.weight", "backbone_net.model._blocks.25._bn1.bias", "backbone_net.model._blocks.25._bn1.running_mean", "backbone_net.model._blocks.25._bn1.running_var", "backbone_net.model._blocks.25._se_reduce.conv.weight", "backbone_net.model._blocks.25._se_reduce.conv.bias", "backbone_net.model._blocks.25._se_expand.conv.weight", "backbone_net.model._blocks.25._se_expand.conv.bias", "backbone_net.model._blocks.25._project_conv.conv.weight", "backbone_net.model._blocks.25._bn2.weight", "backbone_net.model._blocks.25._bn2.bias", "backbone_net.model._blocks.25._bn2.running_mean", "backbone_net.model._blocks.25._bn2.running_var", "backbone_net.model._blocks.26._expand_conv.conv.weight", "backbone_net.model._blocks.26._bn0.weight", "backbone_net.model._blocks.26._bn0.bias", "backbone_net.model._blocks.26._bn0.running_mean", "backbone_net.model._blocks.26._bn0.running_var", "backbone_net.model._blocks.26._depthwise_conv.conv.weight", "backbone_net.model._blocks.26._bn1.weight", "backbone_net.model._blocks.26._bn1.bias", "backbone_net.model._blocks.26._bn1.running_mean", "backbone_net.model._blocks.26._bn1.running_var", "backbone_net.model._blocks.26._se_reduce.conv.weight", "backbone_net.model._blocks.26._se_reduce.conv.bias", "backbone_net.model._blocks.26._se_expand.conv.weight", "backbone_net.model._blocks.26._se_expand.conv.bias", "backbone_net.model._blocks.26._project_conv.conv.weight", "backbone_net.model._blocks.26._bn2.weight", "backbone_net.model._blocks.26._bn2.bias", "backbone_net.model._blocks.26._bn2.running_mean", "backbone_net.model._blocks.26._bn2.running_var", "backbone_net.model._blocks.27._expand_conv.conv.weight", "backbone_net.model._blocks.27._bn0.weight", "backbone_net.model._blocks.27._bn0.bias", "backbone_net.model._blocks.27._bn0.running_mean", "backbone_net.model._blocks.27._bn0.running_var", "backbone_net.model._blocks.27._depthwise_conv.conv.weight", "backbone_net.model._blocks.27._bn1.weight", "backbone_net.model._blocks.27._bn1.bias", "backbone_net.model._blocks.27._bn1.running_mean", "backbone_net.model._blocks.27._bn1.running_var", "backbone_net.model._blocks.27._se_reduce.conv.weight", "backbone_net.model._blocks.27._se_reduce.conv.bias", "backbone_net.model._blocks.27._se_expand.conv.weight", "backbone_net.model._blocks.27._se_expand.conv.bias", "backbone_net.model._blocks.27._project_conv.conv.weight", "backbone_net.model._blocks.27._bn2.weight", "backbone_net.model._blocks.27._bn2.bias", "backbone_net.model._blocks.27._bn2.running_mean", "backbone_net.model._blocks.27._bn2.running_var", "backbone_net.model._blocks.28._expand_conv.conv.weight", "backbone_net.model._blocks.28._bn0.weight", "backbone_net.model._blocks.28._bn0.bias", "backbone_net.model._blocks.28._bn0.running_mean", "backbone_net.model._blocks.28._bn0.running_var", "backbone_net.model._blocks.28._depthwise_conv.conv.weight", "backbone_net.model._blocks.28._bn1.weight", "backbone_net.model._blocks.28._bn1.bias", "backbone_net.model._blocks.28._bn1.running_mean", "backbone_net.model._blocks.28._bn1.running_var", "backbone_net.model._blocks.28._se_reduce.conv.weight", "backbone_net.model._blocks.28._se_reduce.conv.bias", "backbone_net.model._blocks.28._se_expand.conv.weight", "backbone_net.model._blocks.28._se_expand.conv.bias", "backbone_net.model._blocks.28._project_conv.conv.weight", "backbone_net.model._blocks.28._bn2.weight", "backbone_net.model._blocks.28._bn2.bias", "backbone_net.model._blocks.28._bn2.running_mean", "backbone_net.model._blocks.28._bn2.running_var", "backbone_net.model._blocks.29._expand_conv.conv.weight", "backbone_net.model._blocks.29._bn0.weight", "backbone_net.model._blocks.29._bn0.bias", "backbone_net.model._blocks.29._bn0.running_mean", "backbone_net.model._blocks.29._bn0.running_var", "backbone_net.model._blocks.29._depthwise_conv.conv.weight", "backbone_net.model._blocks.29._bn1.weight", "backbone_net.model._blocks.29._bn1.bias", "backbone_net.model._blocks.29._bn1.running_mean", "backbone_net.model._blocks.29._bn1.running_var", "backbone_net.model._blocks.29._se_reduce.conv.weight", "backbone_net.model._blocks.29._se_reduce.conv.bias", "backbone_net.model._blocks.29._se_expand.conv.weight", "backbone_net.model._blocks.29._se_expand.conv.bias", "backbone_net.model._blocks.29._project_conv.conv.weight", "backbone_net.model._blocks.29._bn2.weight", "backbone_net.model._blocks.29._bn2.bias", "backbone_net.model._blocks.29._bn2.running_mean", "backbone_net.model._blocks.29._bn2.running_var", "backbone_net.model._blocks.30._expand_conv.conv.weight", "backbone_net.model._blocks.30._bn0.weight", "backbone_net.model._blocks.30._bn0.bias", "backbone_net.model._blocks.30._bn0.running_mean", "backbone_net.model._blocks.30._bn0.running_var", "backbone_net.model._blocks.30._depthwise_conv.conv.weight", "backbone_net.model._blocks.30._bn1.weight", "backbone_net.model._blocks.30._bn1.bias", "backbone_net.model._blocks.30._bn1.running_mean", "backbone_net.model._blocks.30._bn1.running_var", "backbone_net.model._blocks.30._se_reduce.conv.weight", "backbone_net.model._blocks.30._se_reduce.conv.bias", "backbone_net.model._blocks.30._se_expand.conv.weight", "backbone_net.model._blocks.30._se_expand.conv.bias", "backbone_net.model._blocks.30._project_conv.conv.weight", "backbone_net.model._blocks.30._bn2.weight", "backbone_net.model._blocks.30._bn2.bias", "backbone_net.model._blocks.30._bn2.running_mean", "backbone_net.model._blocks.30._bn2.running_var", "backbone_net.model._blocks.31._expand_conv.conv.weight", "backbone_net.model._blocks.31._bn0.weight", "backbone_net.model._blocks.31._bn0.bias", "backbone_net.model._blocks.31._bn0.running_mean", "backbone_net.model._blocks.31._bn0.running_var", "backbone_net.model._blocks.31._depthwise_conv.conv.weight", "backbone_net.model._blocks.31._bn1.weight", "backbone_net.model._blocks.31._bn1.bias", "backbone_net.model._blocks.31._bn1.running_mean", "backbone_net.model._blocks.31._bn1.running_var", "backbone_net.model._blocks.31._se_reduce.conv.weight", "backbone_net.model._blocks.31._se_reduce.conv.bias", "backbone_net.model._blocks.31._se_expand.conv.weight", "backbone_net.model._blocks.31._se_expand.conv.bias", "backbone_net.model._blocks.31._project_conv.conv.weight", "backbone_net.model._blocks.31._bn2.weight", "backbone_net.model._blocks.31._bn2.bias", "backbone_net.model._blocks.31._bn2.running_mean", "backbone_net.model._blocks.31._bn2.running_var", "backbone_net.model._blocks.32._expand_conv.conv.weight", "backbone_net.model._blocks.32._bn0.weight", "backbone_net.model._blocks.32._bn0.bias", "backbone_net.model._blocks.32._bn0.running_mean", "backbone_net.model._blocks.32._bn0.running_var", "backbone_net.model._blocks.32._depthwise_conv.conv.weight", "backbone_net.model._blocks.32._bn1.weight", "backbone_net.model._blocks.32._bn1.bias", "backbone_net.model._blocks.32._bn1.running_mean", "backbone_net.model._blocks.32._bn1.running_var", "backbone_net.model._blocks.32._se_reduce.conv.weight", "backbone_net.model._blocks.32._se_reduce.conv.bias", "backbone_net.model._blocks.32._se_expand.conv.weight", "backbone_net.model._blocks.32._se_expand.conv.bias", "backbone_net.model._blocks.32._project_conv.conv.weight", "backbone_net.model._blocks.32._bn2.weight", "backbone_net.model._blocks.32._bn2.bias", "backbone_net.model._blocks.32._bn2.running_mean", "backbone_net.model._blocks.32._bn2.running_var", "backbone_net.model._blocks.33._expand_conv.conv.weight", "backbone_net.model._blocks.33._bn0.weight", "backbone_net.model._blocks.33._bn0.bias", "backbone_net.model._blocks.33._bn0.running_mean", "backbone_net.model._blocks.33._bn0.running_var", "backbone_net.model._blocks.33._depthwise_conv.conv.weight", "backbone_net.model._blocks.33._bn1.weight", "backbone_net.model._blocks.33._bn1.bias", "backbone_net.model._blocks.33._bn1.running_mean", "backbone_net.model._blocks.33._bn1.running_var", "backbone_net.model._blocks.33._se_reduce.conv.weight", "backbone_net.model._blocks.33._se_reduce.conv.bias", "backbone_net.model._blocks.33._se_expand.conv.weight", "backbone_net.model._blocks.33._se_expand.conv.bias", "backbone_net.model._blocks.33._project_conv.conv.weight", "backbone_net.model._blocks.33._bn2.weight", "backbone_net.model._blocks.33._bn2.bias", "backbone_net.model._blocks.33._bn2.running_mean", "backbone_net.model._blocks.33._bn2.running_var", "backbone_net.model._blocks.34._expand_conv.conv.weight", "backbone_net.model._blocks.34._bn0.weight", "backbone_net.model._blocks.34._bn0.bias", "backbone_net.model._blocks.34._bn0.running_mean", "backbone_net.model._blocks.34._bn0.running_var", "backbone_net.model._blocks.34._depthwise_conv.conv.weight", "backbone_net.model._blocks.34._bn1.weight", "backbone_net.model._blocks.34._bn1.bias", "backbone_net.model._blocks.34._bn1.running_mean", "backbone_net.model._blocks.34._bn1.running_var", "backbone_net.model._blocks.34._se_reduce.conv.weight", "backbone_net.model._blocks.34._se_reduce.conv.bias", "backbone_net.model._blocks.34._se_expand.conv.weight", "backbone_net.model._blocks.34._se_expand.conv.bias", "backbone_net.model._blocks.34._project_conv.conv.weight", "backbone_net.model._blocks.34._bn2.weight", "backbone_net.model._blocks.34._bn2.bias", "backbone_net.model._blocks.34._bn2.running_mean", "backbone_net.model._blocks.34._bn2.running_var", "backbone_net.model._blocks.35._expand_conv.conv.weight", "backbone_net.model._blocks.35._bn0.weight", "backbone_net.model._blocks.35._bn0.bias", "backbone_net.model._blocks.35._bn0.running_mean", "backbone_net.model._blocks.35._bn0.running_var", "backbone_net.model._blocks.35._depthwise_conv.conv.weight", "backbone_net.model._blocks.35._bn1.weight", "backbone_net.model._blocks.35._bn1.bias", "backbone_net.model._blocks.35._bn1.running_mean", "backbone_net.model._blocks.35._bn1.running_var", "backbone_net.model._blocks.35._se_reduce.conv.weight", "backbone_net.model._blocks.35._se_reduce.conv.bias", "backbone_net.model._blocks.35._se_expand.conv.weight", "backbone_net.model._blocks.35._se_expand.conv.bias", "backbone_net.model._blocks.35._project_conv.conv.weight", "backbone_net.model._blocks.35._bn2.weight", "backbone_net.model._blocks.35._bn2.bias", "backbone_net.model._blocks.35._bn2.running_mean", "backbone_net.model._blocks.35._bn2.running_var", "backbone_net.model._blocks.36._expand_conv.conv.weight", "backbone_net.model._blocks.36._bn0.weight", "backbone_net.model._blocks.36._bn0.bias", "backbone_net.model._blocks.36._bn0.running_mean", "backbone_net.model._blocks.36._bn0.running_var", "backbone_net.model._blocks.36._depthwise_conv.conv.weight", "backbone_net.model._blocks.36._bn1.weight", "backbone_net.model._blocks.36._bn1.bias", "backbone_net.model._blocks.36._bn1.running_mean", "backbone_net.model._blocks.36._bn1.running_var", "backbone_net.model._blocks.36._se_reduce.conv.weight", "backbone_net.model._blocks.36._se_reduce.conv.bias", "backbone_net.model._blocks.36._se_expand.conv.weight", "backbone_net.model._blocks.36._se_expand.conv.bias", "backbone_net.model._blocks.36._project_conv.conv.weight", "backbone_net.model._blocks.36._bn2.weight", "backbone_net.model._blocks.36._bn2.bias", "backbone_net.model._blocks.36._bn2.running_mean", "backbone_net.model._blocks.36._bn2.running_var", "backbone_net.model._blocks.37._expand_conv.conv.weight", "backbone_net.model._blocks.37._bn0.weight", "backbone_net.model._blocks.37._bn0.bias", "backbone_net.model._blocks.37._bn0.running_mean", "backbone_net.model._blocks.37._bn0.running_var", "backbone_net.model._blocks.37._depthwise_conv.conv.weight", "backbone_net.model._blocks.37._bn1.weight", "backbone_net.model._blocks.37._bn1.bias", "backbone_net.model._blocks.37._bn1.running_mean", "backbone_net.model._blocks.37._bn1.running_var", "backbone_net.model._blocks.37._se_reduce.conv.weight", "backbone_net.model._blocks.37._se_reduce.conv.bias", "backbone_net.model._blocks.37._se_expand.conv.weight", "backbone_net.model._blocks.37._se_expand.conv.bias", "backbone_net.model._blocks.37._project_conv.conv.weight", "backbone_net.model._blocks.37._bn2.weight", "backbone_net.model._blocks.37._bn2.bias", "backbone_net.model._blocks.37._bn2.running_mean", "backbone_net.model._blocks.37._bn2.running_var", "backbone_net.model._blocks.38._expand_conv.conv.weight", "backbone_net.model._blocks.38._bn0.weight", "backbone_net.model._blocks.38._bn0.bias", "backbone_net.model._blocks.38._bn0.running_mean", "backbone_net.model._blocks.38._bn0.running_var", "backbone_net.model._blocks.38._depthwise_conv.conv.weight", "backbone_net.model._blocks.38._bn1.weight", "backbone_net.model._blocks.38._bn1.bias", "backbone_net.model._blocks.38._bn1.running_mean", "backbone_net.model._blocks.38._bn1.running_var", "backbone_net.model._blocks.38._se_reduce.conv.weight", "backbone_net.model._blocks.38._se_reduce.conv.bias", "backbone_net.model._blocks.38._se_expand.conv.weight", "backbone_net.model._blocks.38._se_expand.conv.bias", "backbone_net.model._blocks.38._project_conv.conv.weight", "backbone_net.model._blocks.38._bn2.weight", "backbone_net.model._blocks.38._bn2.bias", "backbone_net.model._blocks.38._bn2.running_mean", "backbone_net.model._blocks.38._bn2.running_var". Unexpected key(s) in state_dict: "backbone_net.model._blocks.2._expand_conv.conv.weight", "backbone_net.model._blocks.2._bn0.weight", "backbone_net.model._blocks.2._bn0.bias", "backbone_net.model._blocks.2._bn0.running_mean", "backbone_net.model._blocks.2._bn0.running_var", "backbone_net.model._blocks.2._bn0.num_batches_tracked". size mismatch for bifpn.0.conv6_up.depthwise_conv.conv.weight: copying a param with shape torch.Size([112, 1, 3, 3]) from checkpoint, the shape in current model is torch.Size([288, 1, 3, 3]). size mismatch for bifpn.0.conv6_up.pointwise_conv.conv.weight: copying a param with shape torch.Size([112, 112, 1, 1]) from checkpoint, the shape in current model is torch.Size([288, 288, 1, 1]). size mismatch for bifpn.0.conv6_up.pointwise_conv.conv.bias: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv6_up.bn.weight: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv6_up.bn.bias: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv6_up.bn.running_mean: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv6_up.bn.running_var: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv5_up.depthwise_conv.conv.weight: copying a param with shape torch.Size([112, 1, 3, 3]) from checkpoint, the shape in current model is torch.Size([288, 1, 3, 3]). size mismatch for bifpn.0.conv5_up.pointwise_conv.conv.weight: copying a param with shape torch.Size([112, 112, 1, 1]) from checkpoint, the shape in current model is torch.Size([288, 288, 1, 1]). size mismatch for bifpn.0.conv5_up.pointwise_conv.conv.bias: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv5_up.bn.weight: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv5_up.bn.bias: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv5_up.bn.running_mean: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv5_up.bn.running_var: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv4_up.depthwise_conv.conv.weight: copying a param with shape torch.Size([112, 1, 3, 3]) from checkpoint, the shape in current model is torch.Size([288, 1, 3, 3]). size mismatch for bifpn.0.conv4_up.pointwise_conv.conv.weight: copying a param with shape torch.Size([112, 112, 1, 1]) from checkpoint, the shape in current model is torch.Size([288, 288, 1, 1]). size mismatch for bifpn.0.conv4_up.pointwise_conv.conv.bias: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]). size mismatch for bifpn.0.conv4_up.bn.weight: copying a param with shape torch.Size([112]) from checkpoint, the shape in current model is torch.Size([288]).
..so, I think I have to train my own weights. But is it normal, that my Regression Loss stays literally zero?
can you share your training command and your project file
!pip install pycocotools numpy opencv-python tqdm tensorboard tensorboardX pyyaml webcolors
!pip install torch==1.4.0
!pip install torchvision==0.5.0
from google.colab import drive
drive.mount('/content/drive')
!python train.py -c 0 -p project_malte --head_only True --lr 1e-3 --batch_size 32 --load_weights weights/efficientdet-d0.pth --num_epochs 50 --save_interval 100
#get latest weight file
%cd logs/project_malte
weight_file = !ls -Art | grep efficientdet
%cd ../..
#uncomment the next line to specify a weight file
weight_file[-1] = 'efficientdet-d1_499_18500.pth'
%cd /content/drive/MyDrive/Colab Notebooks/Yet-Another-EfficientDet-Pytorch
!python coco_eval.py -c 1 -p project_malte -w "logs/project_malte/{weight_file[-1]}"
It did work - but still no Regression loss and somehow size-format errors
Why your weights is efficientdet d1 while you trained d0. I think you loaded the wrong weights
I trained with self-trained, d0, d1, d2 - the screenshots are not subsequently.
I tried out different weights, but neither of them solved the Regression loss error
I see, coco_eval.py also need to specify the co-efficient, like -c 1
, 0 by default. So you never really loaded your weights
About that 0 reg loss, can you visualize your annotations by drawing boxes on your images
Image and labels look like this::
ok, I'll take a look first
Thank you so much!
Found it, category id should start from 1, instead of 0.
oh, wow! I will try it out immediately! this is awesome, thank you so much
Hi, when starting to train it produces the following error:
Error] Traceback (most recent call last): File "train.py", line 226, in train cls_loss, reg_loss = model(imgs, annot, obj_list=params.obj_list) File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 532, in call result = self.forward(*input, **kwargs) File "/usr/local/lib/python3.6/dist-packages/torch/nn/parallel/data_parallel.py", line 148, in forward inputs, kwargs = self.scatter(inputs, kwargs, self.device_ids) File "/content/drive/My Drive/Colab Notebooks/Yet-Another-EfficientDet-Pytorch/utils/utils.py", line 205, in scatter for device_idx in range(len(devices))], \ File "/content/drive/My Drive/Colab Notebooks/Yet-Another-EfficientDet-Pytorch/utils/utils.py", line 205, in
for device_idx in range(len(devices))], \
RuntimeError: CUDA error: invalid device ordinal