When running extract-feature.py --gpus=0,1,2,3,4 ... on multi-GPU environment,
recognition result by GPU-0 is fair, but results by GPU-1,2,3,4 are not fair.
For instance, all softmax values are 0.0025.
I think that the script loads model to the first GPU only.
When running
extract-feature.py --gpus=0,1,2,3,4 ...
on multi-GPU environment, recognition result by GPU-0 is fair, but results by GPU-1,2,3,4 are not fair. For instance, all softmax values are 0.0025.I think that the script loads model to the first GPU only.