search

Ninebell / GaneratedHandsForReal_TIME

it referenced paper of GANerated Hands for Real-Time 3D Hand Tracking from Monocular RGB.
MIT License
86 stars 17 forks source link

The silnet's output mask is not a binary mask. but more like a detected edge!!! #14

Open visonpon opened 4 years ago

Ninebell commented 4 years ago

sorry, sir what did you mean?? Did you mean silnet output is image of edge, not binary mask?

visonpon commented 4 years ago

@Ninebell yeah, I tested the trained model you provided on my own image, and get the strange result, hope you can give some advice, thanks~

Ninebell commented 4 years ago

@visonpon can you show me some examples?

Ninebell commented 4 years ago

@visonpon trained silnet model can work image which background is white

visonpon commented 4 years ago

1938378466 oh, my test images' background is green, maybe this is why I got bad results. btw, I have trained silnet on my own data, but at the beginning of the training phase, the accuracy has achieved to 80+, I also curious about this...

Ninebell commented 4 years ago

@visonpon Is dataset personal?? or Just your dataset?