The problem with testing coco datasets
Closed this issue · 2 comments
HMCCMH commented
When the code finishing the test, the program was killed. And the detection.pkl was 0 Kb and there is no results.
INFO test_engine.py: 281: im_detect: range [1, 40504] of 40504: 40471/40504 1.306s (eta: 0:00:43)
INFO test_engine.py: 281: im_detect: range [1, 40504] of 40504: 40481/40504 1.306s (eta: 0:00:30)
INFO test_engine.py: 281: im_detect: range [1, 40504] of 40504: 40491/40504 1.306s (eta: 0:00:16)
INFO test_engine.py: 281: im_detect: range [1, 40504] of 40504: 40501/40504 1.306s (eta: 0:00:03)
Killed
bityangke commented
OOM？
HMCCMH commented
OOM？
I tested it agin and used dmesg | egrep -i -B100 'killed process'. And I found that it was OOM.
[12686762.082512] Out of memory: Kill process 135509 (python3.6) score 894 or sacrifice child
[12686762.082887] Killed process 135509 (python3.6) total-vm:324591472kB, anon-rss:233040056kB, file-rss:85428kB
So the detections.pkl can't be saved.