search

facebookresearch / DensePose

A real-time approach for mapping all human pixels of 2D RGB images to a 3D surface-based model of the body
http://densepose.org
Other
6.98k stars 1.3k forks source link

Test Accuracy Compared to Paper #126

Closed landersson closed 6 years ago

landersson commented 6 years ago

Hi,

Running the following test_net.py commandline:

python2 tools/test_net.py --cfg configs/DensePose_ResNet50_FPN_s1x-e2e.yaml TEST.WEIGHTS https://s3.amazonaws.com/densepose/DensePose_ResNet50_FPN_s1x-e2e.pkl NUM_GPUS 1

I get the following body_uv accuracy:

INFO task_evaluation.py: 200: copypaste: 0.4892,0.8490,0.5078,0.4384,0.5059

The overall AP that I get, 0.4892, seems quite different from the one reported on the first row of Table 1 in the paper, i.e 51.0.

Would you expect them to be more similar?

WongKinYiu commented 6 years ago

"Due to the new per-part normalization the AP numbers do not match those reported in the paper, which are obtained with global normalization factor K = 0.255."

landersson commented 6 years ago

Ok, thanks, that explains it. I'll close this issue.