Closed HanwenCao closed 4 years ago
Hi Hanwen, the pickle file is the weights of the Mask R-CNN detector from detectron2. The slow downloading is probably caused by the GFW, so using some proxy may improve it. You may also find another way to manually download the pkl
file, put it in some directory, and set environment variable FVCORE_CACHE
to that directory.
Hi Hanwen, the pickle file is the weights of the Mask R-CNN detector from detectron2. The slow downloading is probably caused by the GFW, so using some proxy may improve it. You may also find another way to manually download the
pkl
file, put it in some directory, and set environment variableFVCORE_CACHE
to that directory.
Hi Lijun,
Thanks for your very quick reply. The second solution seems good. I found the download link :
https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x/138205316/model_final_a3ec72.pkl
which seems to match the one in your implementation.
However, I am SUPER new to detectron2. Could you please explain how I should set environment variable FVCORE_CACHE
?
Yep, that should be the correct file. Say you put that file at /data/model_final_a3ec72.pkl
, then you should set FVCORE_CACHE
to /data
.
Run by FVCORE_CACHE=/data ./test.bash
should be enough. You can also try setting it in python via os.environ['FVCORE_CACHE'] = '/data'
at the top line.
Yep, that should be the correct file. Say you put that file at
/data/model_final_a3ec72.pkl
, then you should setFVCORE_CACHE
to/data
. Run byFVCORE_CACHE=/data ./test.bash
should be enough. You can also try setting it in python viaos.environ['FVCORE_CACHE'] = '/data'
at the top line.
Thank you so much. But it did not work. I am not sure if I am doing right. Here is what I did.
/data/caohw9/track3_torch/model_final_a3ec72.pkl
FVCORE_CACHE=/data/caohw9/track3_torch
bash test.sh
. It did not work so I added os.environ['FVCORE_CACHE'] = '/data/caohw9/track3_torch'
to the top of utils/test.py But it still did not work (still tried to download model_final_a3ec72.pkl). I was wondering if I made some mistakes?
I added a line at the top of test.sh.
FVCORE_CACHE=/data/caohw9/track3_torch
You should use export FVCORE_CACHE=/data/caohw9/track3_torch
, or the variable won't be passed into subprocesses.
I added a line at the top of test.sh.
FVCORE_CACHE=/data/caohw9/track3_torch
You should use
export FVCORE_CACHE=/data/caohw9/track3_torch
, or the variable won't be passed into subprocesses.
Hi Lijun,
Thank you for your explanation. I just found that os.environ['FVCORE_CACHE'] = 'SOMEPATH'
is helpful. However, the pkl file should under a specific structure:
SOMEPATH/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x/138205316/FILE.pkl
Thanks!
Hi! I tried to run "test.bash" and it used too much time to download "model_final_a3ec72.pkl". Although it did not succeeded because of the timeout error (Wait time out after 600 seconds). I was wondering what is this pkl file for? Could I bypass the slow downloading? Thank you!
2020-07-25 14:22:52,011 __main__ INFO Testing with args: ['/data/caohw9/track3_dataset/test/S06', '/code/ELECTRICITY-MTMC/exp/file-list.txt', '/code/ELECTRICITY-MTMC/exp/tracklets.txt'] 2020-07-25 14:22:52,013 ELECTRICITY-MTMC.utils.run INFO Running with args: Namespace(cache_dir=None, dataset_dir='/data/caohw9/track3_dataset/test/S06', expand=0, frame_timeout=100, init_timeout=600, min_track_length=20, silent=False, stride=1, system_output='/code/ELECTRICITY-MTMC/exp/tracklets.txt', video_list_file='/code/ELECTRICITY-MTMC/exp/file-list.txt') 2020-07-25 14:22:55,239 VideoSystem WARNING 4 gpus requested, but only 1 gpus (gpu id: [0]) available 2020-07-25 14:22:55,239 ELECTRICITY-MTMC.utils.run INFO Running 6 jobs 2020-07-25 14:22:55,241 VideoSystem INFO System initializing ... model_final_a3ec72.pkl: 13%|#6 | 33.1M/254M [09:47<1:45:34, 34.9kB/s]