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
6.94k stars 1.3k forks source link

out of memory Error from operator? #100

Open loongofqiao opened 6 years ago

loongofqiao commented 6 years ago

i want to train a model that detecting keypoints,i get a error that is out of memory when i run the command"python tools/train_net.py --cfg configs/DensePose_ResNet50_FPN_single_GPU.yaml OUTPUT_DIR /tmp//trained-model", and i have configured that yaml file, i add the KRCNN configuration. and my GPU is GTX1060

vkhalidov commented 6 years ago

You can try reducing TRAIN.IMS_PER_BATCH to 1 in your config file

loongofqiao commented 6 years ago

@vkhalidov Thanks for your reply! I have changed the batch size, i worked. And i have an other question, i realized NUM_KEYPOINTS is 17 in my yaml file, because eg.. 17 for coco dataset! but densepose coco dataset have about 100 keypoints, so can i chang the NUM_KEYPOINTS to eg.. 50?

vkhalidov commented 6 years ago

These annotations are different. I would not call densepose annotations "keypoints" since these points are sampled randomly and do not correspond to the same body locations.

loongofqiao commented 6 years ago

@vkhalidov I see，thank you very much！

JustinhoCHN commented 5 years ago

@vkhalidov Excuse me, my graphic card is gtx1080ti with 11GB memory, and I reduced TRAIN.IMS_PER_BATCH to 1, but still got that error, why?