RuntimeError: CUDA error: an illegal memory access was encountered

[sdimantsd]

Hi,
I trained my own dataset using yolact edge trainer.
When I try to eval I get this error:

python3 eval.py --use_fp16_tensorrt --trained_model=./weights/yolact_nets_shuk_resnet101_im320_9_130000.pth --images=/home/ws/images/imgs_in/:/home/ws/imgs_out --top_k=20 --score_threshold=0.3
Config not specified. Parsed yolact_nets_shuk_resnet101_im320_config from the file name.

[01/31 09:54:40 yolact.eval]: Loading model...
[01/31 09:54:50 yolact.eval]: Model loaded.
[01/31 09:54:50 yolact.eval]: Converting to TensorRT...
[01/31 09:54:50 yolact.eval]: Converting backbone to TensorRT...
[01/31 09:55:35 yolact.eval]: Converting protonet to TensorRT...
[01/31 09:55:41 yolact.eval]: Converting FPN to TensorRT...
Warning: Encountered known unsupported method torch.zeros
[01/31 09:55:51 yolact.eval]: Converting PredictionModule to TensorRT...
[01/31 09:56:02 yolact.eval]: Converted to TensorRT.

WARNING [01/31 09:56:03 yolact.layers.detect]: Encountered IndexError as mentioned in https://github.com/haotian-liu/yolact_edge/issues/27. Flattening predictions to avoid error, please verify the outputs. If there are any problems you met related to this, please report an issue.
/home/ws/images/imgs_in/13.jpg -> /home/ws/imgs_out/13.png
Traceback (most recent call last):
  File "eval.py", line 1246, in <module>
    evaluate(net, dataset)
  File "eval.py", line 887, in evaluate
    evalimages(net, inp, out)
  File "eval.py", line 619, in evalimages
    evalimage(net, path, out_path)
  File "eval.py", line 594, in evalimage
    preds = net(batch, extras=extras)["pred_outs"]
  File "/home/ws/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/ws/DL/yolact_edge/yolact.py", line 1879, in forward
    outs_wrapper["pred_outs"] = self.detect(pred_outs, extras=extras)
  File "/home/ws/DL/yolact_edge/layers/functions/detection.py", line 77, in __call__
    decoded_boxes = decode(loc_data[batch_idx], prior_data)
RuntimeError: The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
  File "/home/ws/DL/yolact_edge/layers/box_utils.py", line 264, in decode
        boxes = torch.cat((
            priors[:, :2] + loc[:, :2] * variances[0] * priors[:, 2:],
            priors[:, 2:] * torch.exp(loc[:, 2:] * variances[1])), 1)
            ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
        boxes[:, :2] -= boxes[:, 2:] / 2
        boxes[:, 2:] += boxes[:, :2]
RuntimeError: CUDA error: an illegal memory access was encountered

If I use --disable_tensorrt, it's works without any error.

[haotian-liu]

It seems that for some of the models, the TensorRT conversion will cause some of the issues. Investigating the issue as in #38 and #42.

[sdimantsd]

Oh OK.
Thanks :-)

[haotian-liu]

Hi I am currently closing this issue, and merge the discussion related to TensorRT conversion issue after training on a custom dataset to this issue #47 as it is quite hard for me to track so many open issues. Hope you understand, thanks.

[haotian-liu]

I somehow figured out that the cause and applied the fix, details of the solution are explained in #47. Please take a look to see if the issue can be resolved.
If the issue persists, please reply directly to #47 (this will be the main thread to deal with related issues for now) with experiment configurations (details also explain there). Thanks.

[sdimantsd]

Thanks