pytorch/examples

Why there is no tanh activation at the end of TransformerNet?

Open

#570 opened on Jun 3, 2019

View on GitHub
 (0 comments) (0 reactions) (0 assignees)Python (21,634 stars) (9,429 forks)batch import
good first issue

Description

In Fast Style Transfer, it seems like there is no tanh layer at the end of TransformerNet model. There is no gurantee that output values fti in certain range e.g. [0, 1] for PIL image.

Using only deconv is also valid? How does it work?

Contributor guide

Tech stack
python
Domain
machine learning
Issue type
research
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
basic understanding of style transferknowledge of activation functions
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
30
Research direction
Investigate the TransformerNet implementation in the pytorch/examples repository. Review the model definition (likely in fast neural style.py) to confirm the absence of a tanh activation. Examine the output range of the deconvolution layers and understand how the image is scaled to [0,1] in the training or inference pipeline. Look for any normalization applied after the network.