Some problem when evaluate the model performance

[yklilfft]

Thanks a lot for open source!
I 've got some problems during test model performance by using ./script.sh.
Simply run ./script.sh to evaluate the model trained on THUMOS14 , error happens:
ValueError: invalid literal for int{} with base 10: '1]]]'

I checked the test_log file, find a weird data:
frames:[[0 0 0 1]]
not like others format:
frames:[[ 0 0 768 1]] ( e.g.)
then I rewrite the function get_segments() in thumos14_log_analysis.py to fix this bug.

However, the mAP is very very low with about 0.003 mAP.

I tried model u provided to test again , the mAP is very low lilke that above.

I wanna know what happened? I just simply run ur train code and test code without any change