ymzis69/alpharefine_tensorrt

This repo only includes tensorRT version of AlphaRefine module, not including other base trackers

alpharefine_tensorrt's Issues

AttributeError: 'function' object has no attribute 'named_modules'

/usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/TensorBody.h:354:7: note: declared here
T * data() const {
^~~~
[3/3] c++ prroi_pooling_gpu.o prroi_pooling_gpu_impl.cuda.o -shared -L/usr/local/lib/python3.6/dist-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o _prroi_pooling.so
Loading extension module _prroi_pooling...
0%| | 1/3085 [00:40<34:37:39, 40.42s/it]
Traceback (most recent call last):
File "./arena/LaSOT/run_RF_RF.py", line 118, in
main()
File "./arena/LaSOT/run_RF_RF.py", line 79, in main
pred_bbox = RF_module.refine(cv2.cvtColor(img, cv2.COLOR_BGR2RGB), np.array(pred_bbox))
File "./pytracking/refine_modules/refine_module.py", line 60, in refine
output = self.refine_network.forward_test(Cpatch_tensor, mode='test', convert_trt=self.convert_trt) # (1,1,H,W)
File "./ltr/models/SEx_beta/SEcm_mbv2.py", line 86, in forward_test
backbone_mbv2_256_trt = torch2trt(self.extract_backbone_features, [test_imgs.view(-1, *test_imgs.shape[-3:])])
File "/usr/local/lib/python3.6/dist-packages/torch2trt-0.4.0-py3.6.egg/torch2trt/torch2trt.py", line 773, in torch2trt
with ConversionContext(network, torch2trt_kwargs=kwargs, builder_config=config, logger=logger) as ctx:
File "/usr/local/lib/python3.6/dist-packages/torch2trt-0.4.0-py3.6.egg/torch2trt/torch2trt.py", line 436, in init
for name, module in torch2trt_kwargs['module'].named_modules():
AttributeError: 'function' object has no attribute 'named_modules'

大神求帮助看看问题出在哪里呀。我的torch torchvision torch2trt版本和你一模一样

How to run tensorRT version with ARnet model

hi
I ran the non-tensorRT version the "ARnet_seg_mask_ep0040.pth.tar" model and now I'm going to run the tensorRT version with this model. How to run the AlphaRefine algorithm of the tensorRT version with the "ARnet_seg_mask_ep0040.pth.tar" model. Thank you for guiding me.

Need help converting the model to trt

Hi,
I'm having problems converting resnet34 (SEcmnet_ep0040-c.pth.tar) to .trt.
I downloaded the pth.tar and place it the indicated folder and when I tried to run the suggested command

python ./arena/LaSOT/run_RF_RF.py --tracker_name siamrpn_r50_l234_dwxcorr --dataset UAV123 --convert_trt

I get an error that uav123.json doesn't exist.
I search in the repository and find a uav123.json in the folder AlphaRefine_TensorRT/arena/GOT10K/toolkit/got10k/datasets and their fields match with folder in the dataset UAV123 that I can download in its website https://cemse.kaust.edu.sa/ivul/uav123, but I keep getting an error.

The error occur in the file AlphaRefine_TensorRT/pysot/toolkit/datasets/uav.py because it is trying to create a video map with the data from uav123.json

        with open(os.path.join(dataset_root, name+'.json'), 'r') as f:
            meta_data = json.load(f)

        # load videos
        pbar = tqdm(meta_data.keys(), desc='loading '+name, ncols=100)
        self.videos = {}
        for video in pbar:
            pbar.set_postfix_str(video)
            self.videos[video] = UAVVideo(video,
                                          dataset_root,
                                          meta_data[video]['video_dir'],
                                          meta_data[video]['init_rect'],
                                          meta_data[video]['img_names'],
                                          meta_data[video]['gt_rect'],
                                          meta_data[video]['attr'])

uav123.json doesn't have those fields, the structure of the .json is something like this.

{
    "UAV123": {
        "bike1": {
            "start_frame": 1,
            "end_frame": 3085,
            "folder_name": "bike1"
        },
        "bike2": {
            "start_frame": 1,
            "end_frame": 553,
            "folder_name": "bike2"
        },

It looks like the file that should be used is AlphaRefine_TensorRT/arena/GOT10K/toolkit/datasets/uav123.py, that it is in the directory where I found the uva123.json, but run_RF_RF.py calls the one in pysot.

So, can you guys help me? I have no clue how to fix it or what I am doing wrong.
The tutorial doesn't explain either how to use the obtained trt layers in other tracker. I trying to use them to fast dimp50.