About loading weights on docker

[atusi-nakajima]

I would like to load weights on the provided docker, but it is not working, so I have a question. Here is a detailed description of how I did it.

First, I started the container and logged in as root. docker pull deepdrive/spc docker run -it --gpus all deepdrive/spc At this time, the nvidia-smi command worked fine.

I then git cloned spc2 and placed the downloaded weights in /home/carla/spc2/pretrain/. (I renamed the file to model_only_fpn50.pth.)

Enter the spc virtual environment. conda activate spc

Open a separate terminal and start carla. docker exec -it <container ID> bash (spc) carla@325657a8b89c:~/carla_0.8.4$ bash run.sh Confirm that the message "Waiting for the client to connect..." is displayed.

Tried to start training in the previous terminal. (spc) root@325657a8b89c:~/spc2/scripts# bash train_carla.sh

Here are the steps I followed so far. However, I could not start as follows. Please let me know if there is something wrong. Thank you very much.

(spc) root@325657a8b89c:~/spc2/scripts# bash train_carla.sh No model to resume! turn to the base pretrain model: pretrain/model_only_fpn50.pth Traceback (most recent call last): File "main.py", line 95, in <module> main() File "main.py", line 91, in main train_policy(args, env) File "/home/carla/spc2/train.py", line 334, in train_policy trainer = Trainer(args, env) File "/home/carla/spc2/train.py", line 124, in __init__ self.model, self.optim, self.epoch, self.exploration, self.num_steps = init_models(self.args) File "/home/carla/spc2/models/model.py", line 238, in init_models train_net, epoch, step = load_model(args, args.save_path, train_net, resume=args.resume) File "/home/carla/spc2/utils/util.py", line 327, in load_model net.load_state_dict(state_dict) File "/home/carla/anaconda3/envs/spc/lib/python3.5/site-packages/torch/nn/modules/module.py", line 845, in load_state_dict self.__class__.__name__, "\n\t".join(error_msgs))) RuntimeError: Error(s) in loading state_dict for ConvLSTMMulti: Missing key(s) in state_dict: "conv_lstm.feature_map_predictor.identify.bias", "conv_lstm.feature_map_predictor.identify.weight", "conv_lstm.feature_map_predictor.stair.bias", "conv_lstm.feature_map_predictor.stair.weight", "conv_lstm.detector.coll_head.0.bias", "conv_lstm.detector.coll_head.0.weight", "conv_lstm.detector.coll_head.2.bias", "conv_lstm.detector.coll_head.2.weight", "conv_lstm.detector.coll_head.4.bias", "conv_lstm.detector.coll_head.4.weight", "conv_lstm.detector.coll_head.6.bias", "conv_lstm.detector.coll_head.6.weight", "conv_lstm.detector.coll_head.8.bias", "conv_lstm.detector.coll_head.8.weight", "conv_lstm.detector.thrd_head.confidence_head.0.bias", "conv_lstm.detector.thrd_head.confidence_head.0.weight", "conv_lstm.detector.thrd_head.confidence_head.2.bias", "conv_lstm.detector.thrd_head.confidence_head.2.weight", "conv_lstm.detector.thrd_head.confidence_head.4.bias", "conv_lstm.detector.thrd_head.confidence_head.4.weight", "conv_lstm.detector.thrd_head.confidence_head.6.bias", "conv_lstm.detector.thrd_head.confidence_head.6.weight", "conv_lstm.detector.thrd_head.confidence_head.8.bias", "conv_lstm.detector.thrd_head.confidence_head.8.weight", "conv_lstm.detector.thrd_head.residual_head.0.bias", "conv_lstm.detector.thrd_head.residual_head.0.weight", "conv_lstm.detector.thrd_head.residual_head.2.bias", "conv_lstm.detector.thrd_head.residual_head.2.weight", "conv_lstm.detector.thrd_head.residual_head.4.bias", "conv_lstm.detector.thrd_head.residual_head.4.weight", "conv_lstm.detector.thrd_head.residual_head.6.bias", "conv_lstm.detector.thrd_head.residual_head.6.weight", "conv_lstm.detector.thrd_head.residual_head.8.bias", "conv_lstm.detector.thrd_head.residual_head.8.weight", "conv_lstm.detector.thrd_head.dimension_head.0.bias", "conv_lstm.detector.thrd_head.dimension_head.0.weight", "conv_lstm.detector.thrd_head.dimension_head.2.bias", "conv_lstm.detector.thrd_head.dimension_head.2.weight", "conv_lstm.detector.thrd_head.dimension_head.4.bias", "conv_lstm.detector.thrd_head.dimension_head.4.weight", "conv_lstm.detector.thrd_head.dimension_head.6.bias", "conv_lstm.detector.thrd_head.dimension_head.6.weight", "conv_lstm.detector.thrd_head.dimension_head.8.bias", "conv_lstm.detector.thrd_head.dimension_head.8.weight", "conv_lstm.detector.thrd_head.center_head.0.bias", "conv_lstm.detector.thrd_head.center_head.0.weight", "conv_lstm.detector.thrd_head.center_head.2.bias", "conv_lstm.detector.thrd_head.center_head.2.weight", "conv_lstm.detector.thrd_head.center_head.4.bias", "conv_lstm.detector.thrd_head.center_head.4.weight", "conv_lstm.detector.thrd_head.center_head.6.bias", "conv_lstm.detector.thrd_head.center_head.6.weight", "conv_lstm.detector.thrd_head.center_head.8.bias", "conv_lstm.detector.thrd_head.center_head.8.weight", "conv_lstm.depth_head.0.bias", "conv_lstm.depth_head.0.weight", "conv_lstm.depth_head.2.bias", "conv_lstm.depth_head.2.weight", "conv_lstm.depth_head.4.bias", "conv_lstm.depth_head.4.weight", "conv_lstm.depth_head.6.bias", "conv_lstm.depth_head.6.weight", "conv_lstm.guide_layer.fc11.bias", "conv_lstm.guide_layer.fc11.weight", "conv_lstm.guide_layer.fc22.bias", "conv_lstm.guide_layer.fc22.weight", "conv_lstm.guide_layer.fc33.bias", "conv_lstm.guide_layer.fc33.weight", "conv_lstm.coll_layer.fc11.bias", "conv_lstm.coll_layer.fc11.weight", "conv_lstm.coll_layer.fc22.bias", "conv_lstm.coll_layer.fc22.weight", "conv_lstm.coll_layer.fc33.bias", "conv_lstm.coll_layer.fc33.weight", "conv_lstm.offroad_layer.fc11.bias", "conv_lstm.offroad_layer.fc11.weight", "conv_lstm.offroad_layer.fc22.bias", "conv_lstm.offroad_layer.fc22.weight", "conv_lstm.offroad_layer.fc33.bias", "conv_lstm.offroad_layer.fc33.weight", "conv_lstm.offlane_layer.fc11.bias", "conv_lstm.offlane_layer.fc11.weight", "conv_lstm.offlane_layer.fc22.bias", "conv_lstm.offlane_layer.fc22.weight", "conv_lstm.offlane_layer.fc33.bias", "conv_lstm.offlane_layer.fc33.weight", "conv_lstm.speed_layer.fc11.bias", "conv_lstm.speed_layer.fc11.weight", "conv_lstm.speed_layer.fc22.bias", "conv_lstm.speed_layer.fc22.weight", "conv_lstm.speed_layer.fc33.bias", "conv_lstm.speed_layer.fc33.weight". Unexpected key(s) in state_dict: "conv_lstm.detector_pred.loc_head.0.weight", "conv_lstm.detector_pred.loc_head.0.bias", "conv_lstm.detector_pred.loc_head.2.weight", "conv_lstm.detector_pred.loc_head.2.bias", "conv_lstm.detector_pred.loc_head.4.weight", "conv_lstm.detector_pred.loc_head.4.bias", "conv_lstm.detector_pred.loc_head.6.weight", "conv_lstm.detector_pred.loc_head.6.bias", "conv_lstm.detector_pred.loc_head.8.weight", "conv_lstm.detector_pred.loc_head.8.bias", "conv_lstm.detector_pred.cls_head.0.weight", "conv_lstm.detector_pred.cls_head.0.bias", "conv_lstm.detector_pred.cls_head.2.weight", "conv_lstm.detector_pred.cls_head.2.bias", "conv_lstm.detector_pred.cls_head.4.weight", "conv_lstm.detector_pred.cls_head.4.bias", "conv_lstm.detector_pred.cls_head.6.weight", "conv_lstm.detector_pred.cls_head.6.bias", "conv_lstm.detector_pred.cls_head.8.weight", "conv_lstm.detector_pred.cls_head.8.bias", "conv_lstm.coll_vehicle_layer.conv1.weight", "conv_lstm.coll_vehicle_layer.conv1.bias", "conv_lstm.coll_vehicle_layer.conv2.weight", "conv_lstm.coll_vehicle_layer.conv2.bias", "conv_lstm.coll_vehicle_layer.conv3.weight", "conv_lstm.coll_vehicle_layer.conv3.bias", "conv_lstm.coll_vehicle_layer.conv4.weight", "conv_lstm.coll_vehicle_layer.conv4.bias", "conv_lstm.coll_vehicle_layer.fc1.weight", "conv_lstm.coll_vehicle_layer.fc1.bias", "conv_lstm.coll_vehicle_layer.fc2.weight", "conv_lstm.coll_vehicle_layer.fc2.bias", "conv_lstm.coll_vehicle_layer.fc3.weight", "conv_lstm.coll_vehicle_layer.fc3.bias", "conv_lstm.coll_other_layer.conv1.weight", "conv_lstm.coll_other_layer.conv1.bias", "conv_lstm.coll_other_layer.conv2.weight", "conv_lstm.coll_other_layer.conv2.bias", "conv_lstm.coll_other_layer.conv3.weight", "conv_lstm.coll_other_layer.conv3.bias", "conv_lstm.coll_other_layer.conv4.weight", "conv_lstm.coll_other_layer.conv4.bias", "conv_lstm.coll_other_layer.fc1.weight", "conv_lstm.coll_other_layer.fc1.bias", "conv_lstm.coll_other_layer.fc2.weight", "conv_lstm.coll_other_layer.fc2.bias", "conv_lstm.coll_other_layer.fc3.weight", "conv_lstm.coll_other_layer.fc3.bias", "conv_lstm.guide_layer.fc1.weight", "conv_lstm.guide_layer.fc1.bias", "conv_lstm.guide_layer.fc2.weight", "conv_lstm.guide_layer.fc2.bias", "conv_lstm.guide_layer.fc3.weight", "conv_lstm.guide_layer.fc3.bias", "conv_lstm.coll_layer.fc1.weight", "conv_lstm.coll_layer.fc1.bias", "conv_lstm.coll_layer.fc2.weight", "conv_lstm.coll_layer.fc2.bias", "conv_lstm.coll_layer.fc3.weight", "conv_lstm.coll_layer.fc3.bias", "conv_lstm.offroad_layer.fc1.weight", "conv_lstm.offroad_layer.fc1.bias", "conv_lstm.offroad_layer.fc2.weight", "conv_lstm.offroad_layer.fc2.bias", "conv_lstm.offroad_layer.fc3.weight", "conv_lstm.offroad_layer.fc3.bias", "conv_lstm.offlane_layer.fc1.weight", "conv_lstm.offlane_layer.fc1.bias", "conv_lstm.offlane_layer.fc2.weight", "conv_lstm.offlane_layer.fc2.bias", "conv_lstm.offlane_layer.fc3.weight", "conv_lstm.offlane_layer.fc3.bias", "conv_lstm.speed_layer.fc1.weight", "conv_lstm.speed_layer.fc1.bias", "conv_lstm.speed_layer.fc2.weight", "conv_lstm.speed_layer.fc2.bias", "conv_lstm.speed_layer.fc3.weight", "conv_lstm.speed_layer.fc3.bias".

[noahcao]

I checked the uploaded weights and fixed the key naming issues you reported.

The previous naming incompatibility issue is raised because of the version naming shift, I believe the backbone weights in this new version of pertained weights should be fine.

I'm closing this issue as the issue should be fixed now. But please feel free to reopen it or open a new one if you encounter any further issues.

[atusi-nakajima]

Thank you very much. However, I just downloaded the updated readme weight file, and it consists of the following, and I can't find the .pth filel. Do I need to do something to rename the file?

archive
├── data
│   ├── 140239679049472
│   ├── 140239679052976
....................................................
│   └── 140241559087728
├── data.pkl
└── version

[noahcao]

I just checked the link (https://drive.google.com/file/d/1O7u_ifm0HZE1mBCEgnRW7aoyrX7WgEoZ/view?usp=sharing), it should just contain a .pth file. Could you double check your link?

[atusi-nakajima]

The previous file is in zip format, but it will not load. When you unzip it, it will be in a directory format, as described in issues.

[noahcao]

Sorry, I didn't get your point. I checked the history, the previous version should be a "pretrained.pth" file?

But, anyways, how about the current version provided by the link above? Does it work now?

[atusi-nakajima]

I'm sorry for the misunderstanding. I downloaded the link above and it was in zip format. When unzipped, the directory structure looks like this. https://github.com/SysCV/spc2/issues/6#issuecomment-873821869

[noahcao]

That's strange, I triple checked the link: https://drive.google.com/file/d/1O7u_ifm0HZE1mBCEgnRW7aoyrX7WgEoZ/view?usp=sharing. It should just indicate to a .pth file, not a zip file. Have no idea why that happened to you.