Haichao Yu comments

Results 11 comments of


                                            Haichao Yu

How Do I used a COCO Annotated Custom Keypoint Dataset for Custom Objects in MMPose

Hi @jin-s13 , just to make sure I understand it correctly, to train a model on a new dataset with, e.g., 4 key-points in COCO format, one can just 1)...

Questions about UNet architecture and Dice loss

@jakeret Hi, I added a new version of dice loss but did not test them. Sorry about that.

A question on data augmentation part

Hi Liang, May I ask that how many epochs or iterations (batch_size == 1?) of training make the evaluation performance of 72.9 in SAD on val set? Also, what is...

A question on data augmentation part

Thanks!

Does the Binarize() function use STE？

Want to know the same +1

Does the Binarize() function use STE？

Hi @Ashokvardhan , thanks for your insights. I just focused on the the resnet code for cifar10 (resnet_binary.py). In fact, I have two questions about the code. - One is...

Does the Binarize() function use STE？

Regarding the first point: - According to the paper (Equation 4), the gradient of STE is 0 when abs(weight) > 0 and 1 when abs(weight) < 1. If activations are...

Does the Binarize() function use STE？

@itayhubara, Sorry, its my mistake. What about weight STE?

Q4Q4 post-train Resnet-50 cannot get accuracy 73.3.

Hi @ynahshan, any advice? Thanks.

Some questions regarding the code

Hi @csyhhu, Encountered another error [here](https://github.com/csyhhu/MetaQuant/blob/master/meta-quantize.py#L232). `layer.calibration` does not exists. Maybe because `forward` in `net` is never called?