quantization aware training - bugfix for merged weights becoming 0 (typically due...
authorManu Mathew <a0393608@ti.com>
Fri, 27 Mar 2020 12:53:04 +0000 (18:23 +0530)
committerManu Mathew <a0393608@ti.com>
Fri, 27 Mar 2020 12:54:50 +0000 (18:24 +0530)
commitd162e610d91fcdb7f2750182db83f4d46af3c7b8
tree7c20bfa4da7228a53324ab8fc2122b44f7d8e76a
parentab84ec9f9e27a32f45a2e6372f32ba3fe9b1a799
quantization aware training - bugfix for merged weights becoming 0 (typically due to one bn weight becomming 0)
modules/pytorch_jacinto_ai/vision/losses/segmentation_loss.py
modules/pytorch_jacinto_ai/vision/transforms/image_transforms_xv12.py
modules/pytorch_jacinto_ai/xnn/quantize/quant_train_utils.py