dlut-yyc / isomer Goto Github PK

May I ask how long you need to train isomer. It takes 1 hour for me to train one epoch, and 500h for 500 epochs, which is too long

The results of VSOD testing on FBMS dataset with the final model trained in the paper differ a lot from the results given in the paper. May I ask what is the situation? Is the VSOD training method different? mae=0.069 F-MAX=0.825 S-measure=0.840

When I run the code(./scripts/infer.sh) for validation, the code reports the following error:
Traceback (most recent call last):
File "tools/inference.py", line 9, in
from lib.builder import VOSNet
File "/mnt/d/Project/Isomer/tools/../lib/init.py", line 3, in
from .module import *
File "/mnt/d/Project/Isomer/tools/../lib/module/init.py", line 3, in
from .cst import ContextSharingTransformer
File "/mnt/d/Project/Isomer/tools/../lib/module/cst.py", line 7, in
from mmcv.cnn import constant_init, kaiming_init
ImportError: cannot import name 'constant_init' from 'mmcv.cnn' (/home/liao/anaconda3/envs/lsomer/lib/python3.8/site-packages/mmcv/cnn/init.py)

Can you please tell me what the problem is?