search

SysCV / bdd100k-models

Model Zoo of BDD100K Dataset
https://www.bdd100k.com/
Apache License 2.0
300 stars 40 forks source link

Drivable Segmentation Model inference stuck #6

Closed danielzhangau closed 2 years ago

danielzhangau commented 3 years ago

When I am running Deeplabv3+ model by using: python ./test.py configs/drivable/deeplabv3plus_r50-d8_512x1024_40k_drivable_bdd100k.py --format-only --format-dir output It just stuck in around 1490 step image I have tried several different configs, they all have the same issue.

thomasehuang commented 3 years ago

Can you provide your exact environment details so I can try to reproduce the issue?

danielzhangau commented 3 years ago

Hi, here below is my all packages installed under the environment: 

Package           Version
----------------- ---------
addict            2.4.0
certifi           2021.10.8
cycler            0.11.0
fonttools         4.28.2
kiwisolver        1.3.2
matplotlib        3.5.0
mkl-fft           1.3.1
mkl-random        1.2.2
mkl-service       2.4.0
mmcv-full         1.3.18
mmsegmentation    0.19.0
numpy             1.21.2
olefile           0.46
opencv-python     4.5.4.60
packaging         21.3
Pillow            8.4.0
pip               21.3.1
prettytable       2.4.0
pyparsing         3.0.6
python-dateutil   2.8.2
PyYAML            6.0
setuptools        58.0.4
setuptools-scm    6.3.2
six               1.16.0
tomli             1.2.2
torch             1.10.0
torchaudio        0.10.0
torchvision       0.11.1
typing-extensions 3.10.0.2
wcwidth           0.2.5
wheel             0.37.0
yapf              0.31.0
danielzhangau commented 3 years ago

It has the issue when I run this line (this is an example) python test.py configs/drivable/deeplabv3_r50-d8_512x1024_40k_drivable_bdd100k.py --format-only --format-dir output

thomasehuang commented 3 years ago

Hmm, this could either be due to memory issues or the version of mmsegmentation or mmcv. Maybe first monitor the memory usage and see if that's causing problems, since mmsegmentation's inference uses quite a bit. For the packages, for reference I am using mmsegmentation==0.17.0 and mmcv-full==1.3.9. I can run some tests later to see which versions potentially break the code, but maybe if the memory isn't an issue you can try downgrading the packages and seeing if it works?