EMSANet
EMSANet copied to clipboard
TUI-NICR
EMSANet: Efficient Multi-Task RGB-D Scene Analysis for Indoor Environments
Thanks for your excellent work. I read the code carefully. And I notice that when calculating the CE loss and the miou, the class 0 seems to be ignored, why?
I'm trying to run the evaluation on NYUv2 (without pretraining on Hypersim), but I keep receiving this error, I had the problem discussed in https://github.com/TUI-NICR/EMSANet/issues/16 So I commented the line,...
I wonder whether the normal task is working, it is possible to visualize the normal task？
RuntimeError: Error(s) in loading state_dict for EMSANet: Missing key(s) in state_dict: "encoder.backbone_rgb.conv1.weight", "encoder.backbone_rgb.norm1.weight", "encoder.backbone_rgb.norm1.bias", "encoder.backbone_rgb.norm1.running_mean", "encoder.backbone_rgb.norm1.running_var", "encoder.backbone_rgb.layer1.0.conv1_1.weight", "encoder.backbone_rgb.layer1.0.conv1_1.bias", "encoder.backbone_rgb.layer1.0.conv1_2.weight", "encoder.backbone_rgb.layer1.0.norm1.weight", "encoder.backbone_rgb.layer1.0.norm1.bias", "encoder.backbone_rgb.layer1.0.norm1.running_mean", "encoder.backbone_rgb.layer1.0.norm1.running_var", "encoder.backbone_rgb.layer1.0.conv2_1.weight", "encoder.backbone_rgb.layer1.0.conv2_1.bias", "encoder.backbone_rgb.layer1.0.conv2_2.weight", "encoder.backbone_rgb.layer1.0.norm2.weight", "encoder.backbone_rgb.layer1.0.norm2.bias", "encoder.backbone_rgb.layer1.0.norm2.running_mean", .........................................................
File "/content/drive/MyDrive/EMSANet-main/lib/nicr-scene-analysis-datasets/nicr_scene_analysis_datasets/utils/io.py", line 132, in load_creation_metafile with open(filepath) as f: FileNotFoundError: [Errno 2] No such file or directory: './datasets/nyuv2/creation_meta.json'
May I ask what code changes are required to train my own dataset? Thank you very much.
Dear author, I conducted training using the Cityscapes dataset and attempted to visualize the results using the "inference_samples.py" script. However, I encountered an issue with the instance segmentation output, wherein...
I have a question regarding loading a ResNet model other than ResNet-34. I tried to use ResNet-50 with the pre-trained weights provided by PyTorch, but the ResNet-50 model generated from...
Epoch, Task weighting or the other settings, can you show the training command on hypersim ?
Dear authors, Would it be possible to train the model using multi-gpus?
