search

sshaoshuai / PointRCNN

PointRCNN: 3D Object Proposal Generation and Detection from Point Cloud, CVPR 2019.
MIT License
1.71k stars 426 forks source link

RunTime Error when running inference #84

Open fxy2012 opened 5 years ago

fxy2012 commented 5 years ago

Hello, shaoshuai ! I have trained the model follow your steps, but when I ran the following codes, errors occured: python eval_rcnn.py --cfg_file cfgs/default.yaml --ckpt ../output/rpn/default/ckpt/checkpoint_epoch_200.pth --batch_size 4 --eval_mode rcnn

And the error is : RuntimeError: Error(s) in loading state_dict for PointRCNN: Missing key(s) in state_dict: "rcnn_net.SA_modules.0.mlps.0.layer0.conv.weight", "rcnn_net.SA_modules.0.mlps.0.layer0.conv.bias", "rcnn_net.SA_modules.0.mlps.0.layer1.conv.weight", "rcnn_net.SA_modules.0.mlps.0.layer1.conv.bias", "rcnn_net.SA_modules.0.mlps.0.layer2.conv.weight", "rcnn_net.SA_modules.0.mlps.0.layer2.conv.bias", "rcnn_net.SA_modules.1.mlps.0.layer0.conv.weight", "rcnn_net.SA_modules.1.mlps.0.layer0.conv.bias", "rcnn_net.SA_modules.1.mlps.0.layer1.conv.weight", "rcnn_net.SA_modules.1.mlps.0.layer1.conv.bias", "rcnn_net.SA_modules.1.mlps.0.layer2.conv.weight", "rcnn_net.SA_modules.1.mlps.0.layer2.conv.bias", "rcnn_net.SA_modules.2.mlps.0.layer0.conv.weight", "rcnn_net.SA_modules.2.mlps.0.layer0.conv.bias", "rcnn_net.SA_modules.2.mlps.0.layer1.conv.weight", "rcnn_net.SA_modules.2.mlps.0.layer1.conv.bias", "rcnn_net.SA_modules.2.mlps.0.layer2.conv.weight", "rcnn_net.SA_modules.2.mlps.0.layer2.conv.bias", "rcnn_net.xyz_up_layer.layer0.conv.weight", "rcnn_net.xyz_up_layer.layer0.conv.bias", "rcnn_net.xyz_up_layer.layer1.conv.weight", "rcnn_net.xyz_up_layer.layer1.conv.bias", "rcnn_net.merge_down_layer.layer0.conv.weight", "rcnn_net.merge_down_layer.layer0.conv.bias", "rcnn_net.cls_layer.0.conv.weight", "rcnn_net.cls_layer.0.conv.bias", "rcnn_net.cls_layer.2.conv.weight", "rcnn_net.cls_layer.2.conv.bias", "rcnn_net.cls_layer.3.conv.weight", "rcnn_net.cls_layer.3.conv.bias", "rcnn_net.reg_layer.0.conv.weight", "rcnn_net.reg_layer.0.conv.bias", "rcnn_net.reg_layer.2.conv.weight", "rcnn_net.reg_layer.2.conv.bias", "rcnn_net.reg_layer.3.conv.weight", "rcnn_net.reg_layer.3.conv.bias".

bmankirLinker commented 5 years ago

Hello, shaoshuai ! I have trained the model follow your steps, but when I ran the following codes, errors occured: python eval_rcnn.py --cfg_file cfgs/default.yaml --ckpt ../output/rpn/default/ckpt/checkpoint_epoch_200.pth --batch_size 4 --eval_mode rcnn

And the error is : RuntimeError: Error(s) in loading state_dict for PointRCNN: Missing key(s) in state_dict: "rcnn_net.SA_modules.0.mlps.0.layer0.conv.weight", "rcnn_net.SA_modules.0.mlps.0.layer0.conv.bias", "rcnn_net.SA_modules.0.mlps.0.layer1.conv.weight", "rcnn_net.SA_modules.0.mlps.0.layer1.conv.bias", "rcnn_net.SA_modules.0.mlps.0.layer2.conv.weight", "rcnn_net.SA_modules.0.mlps.0.layer2.conv.bias", "rcnn_net.SA_modules.1.mlps.0.layer0.conv.weight", "rcnn_net.SA_modules.1.mlps.0.layer0.conv.bias", "rcnn_net.SA_modules.1.mlps.0.layer1.conv.weight", "rcnn_net.SA_modules.1.mlps.0.layer1.conv.bias", "rcnn_net.SA_modules.1.mlps.0.layer2.conv.weight", "rcnn_net.SA_modules.1.mlps.0.layer2.conv.bias", "rcnn_net.SA_modules.2.mlps.0.layer0.conv.weight", "rcnn_net.SA_modules.2.mlps.0.layer0.conv.bias", "rcnn_net.SA_modules.2.mlps.0.layer1.conv.weight", "rcnn_net.SA_modules.2.mlps.0.layer1.conv.bias", "rcnn_net.SA_modules.2.mlps.0.layer2.conv.weight", "rcnn_net.SA_modules.2.mlps.0.layer2.conv.bias", "rcnn_net.xyz_up_layer.layer0.conv.weight", "rcnn_net.xyz_up_layer.layer0.conv.bias", "rcnn_net.xyz_up_layer.layer1.conv.weight", "rcnn_net.xyz_up_layer.layer1.conv.bias", "rcnn_net.merge_down_layer.layer0.conv.weight", "rcnn_net.merge_down_layer.layer0.conv.bias", "rcnn_net.cls_layer.0.conv.weight", "rcnn_net.cls_layer.0.conv.bias", "rcnn_net.cls_layer.2.conv.weight", "rcnn_net.cls_layer.2.conv.bias", "rcnn_net.cls_layer.3.conv.weight", "rcnn_net.cls_layer.3.conv.bias", "rcnn_net.reg_layer.0.conv.weight", "rcnn_net.reg_layer.0.conv.bias", "rcnn_net.reg_layer.2.conv.weight", "rcnn_net.reg_layer.2.conv.bias", "rcnn_net.reg_layer.3.conv.weight", "rcnn_net.reg_layer.3.conv.bias".

You're trying to evaluate an rpn module on rcnn eval mode. You have to use --eval_mode rpn in your command.

fxy2012 commented 5 years ago

@bmankirLinker , thank you for your reply. But when I use .pth in rcnn , the error still occured. In the last step, I trained 30 epochs as the author listed. When I use the 30th .pth, the error will happened, but when I use others , it will be ok, but the result is not good .

bmankirLinker commented 5 years ago

@bmankirLinker , thank you for your reply. But when I use .pth in rcnn , the error still occured. In the last step, I trained 30 epochs as the author listed. When I use the 30th .pth, the error will happened, but when I use others , it will be ok, but the result is not good .

So when you run --eval-all option, it works fine till 30th epoch then get an error? Never run into such thing, can you share your training commands and if any changes in your local repo?

jviann commented 5 years ago

I had same problem. If you are training with --train_mode rcnn_offline, try using --eval_mode rcnn_offline. For me that solved the issue.

z393 commented 4 years ago

I had same problem. If you are training with --train_mode rcnn_offline, try using --eval_mode rcnn_offline. For me that solved the issue.

Did you solve the problem? I met the same error, but I don't know how to deal it....

jviann commented 4 years ago

I had same problem. If you are training with --train_mode rcnn_offline, try using --eval_mode rcnn_offline. For me that solved the issue.

Did you solve the problem? I met the same error, but I don't know how to deal it....

I had this issue when training/evaluating on offline GT. I solved it by setting --eval_mode flag to rcnn_offline as shown below (just an example):

python eval_rcnn.py --data_path ../data/ --cfg_file cfgs/data.yaml --ckpt ../output/rcnn/data/ckpt/checkpoint_epoch_50.pth --batch_size 1 --eval_mode rcnn_offline --rcnn_eval_roi_dir ../output/rpn/data/eval/epoch_200/val/detections/data --rcnn_eval_feature_dir ../output/rpn/data/eval/epoch_200/val/features