I can't find the resnet network implementation about deep-prior-pp HOT 8 CLOSED

You can check src/net/resnet.py if this is what you are looking for.

from deep-prior-pp.

thanks for your reply very. I can't find where to use ResNet

from deep-prior-pp.

For testing, ResNet is used in e.g. src/test_realtimepipeline.py , for training check the src/main_*_posereg_embeddin.py files, where you might have to replace PoseRegNet with ResNet.

from deep-prior-pp.

thanks for your reply very.I met something confusing after reading the code,in the file main_nyu_posereg_embedding.py line 118
poseNetParams = PoseRegNetParams(type=0, nChan=nChannels, wIn=imgSizeW, hIn=imgSizeH, batchSize=batchSize,numJoints=1, nDims=train_gt3D_embed.shape[1])
why you set numJoints=1 and set nDims = 30 why numJoints not 14?

from deep-prior-pp.

If you read the paper, we use a low-dimensional embedding for the pose. Therefore, the pose is a 1x30 vector and not the 14x3 3D joint locations.

from deep-prior-pp.

sorry.Maybe,I not have fully understand the article. Can you give me some guidance?or some information.emai [email protected] Thx

from deep-prior-pp.

I read some articles about hand pose estimation. The network is relatively shallow. Why not use the existing network? For example, ResNet101. Looking forward to your reply

from deep-prior-pp.

@wangsun
Actually, we use ResNet50, which still runs in realtime.

from deep-prior-pp.