-
Some things we could add:
- [ ] [easy] Use down/right shifted 3x3 conv instead of a right shifted 1x3 conv in the horizontal stack.
- [ ] [medium] Use a U-Net architecture with skip connections (thi…
-
Original PixelCNN paper (https://arxiv.org/pdf/1606.05328.pdf) uses gated unit defined as tanh(a) * sigmoid(b). Same formulation of gated unit is used in the wavenet paper. Yet here you switched to ga…
-
Sik-Ho Tang. [Review: Representation Learning with Contrastive Predictive Coding (CPC/CPCv1)](https://sh-tsang.medium.com/review-representation-learning-with-contrastive-predictive-coding-cpc-cpcv1-8e…
-
NADE
sampled img
validation img
. Because of that, we can't run different example tests …
-
Thank you for the implementation!
I could not find the code/models for samples generated with a prior learned with PixelCNN.
Could you kindly share that? Thanks!
-
Currently, the model does not predict gaussian noise correctly. We assume that the Unet implementation does not work for this case. Thus, find a fitting architecture which successfully fixes this prob…
-
微博内容精选
-
Sik-Ho Tsang. [Review: Representation Learning with Contrastive Predictive Coding (CPC/CPCv1)](https://sh-tsang.medium.com/review-representation-learning-with-contrastive-predictive-coding-cpc-cpcv1-8…
-
When computing metrics for the evaluation/test datasets, the entire set should be processed to get the correct metrics. Our examples often compute metrics with the following pattern:
```python
def…