Manipulate gradients in ulyanov_et_al_2016

[jbueltemeier]

Since in their implementation the forward pass and the backward pass are independent of each other, the backward pass must be manipulated in pytorch to reproduce the results. This includes the removal of gradients automatically created by pytorch, the addition of gradient normalization at certain points, and the addition of gradient normalization.

[codecov[bot]]

Codecov Report

Merging #250 (da2503b) into master (7ba28c6) will decrease coverage by 0.2%. The diff coverage is 77.7%.

@@           Coverage Diff            @@
##           master    #250     +/-   ##
========================================
- Coverage    98.6%   98.4%   -0.3%     
========================================
  Files          39      39             
  Lines        1558    1601     +43     
========================================
+ Hits         1537    1576     +39     
- Misses         21      25      +4     

Impacted Files	Coverage Δ
pystiche_papers/ulyanov_et_al_2016/_loss.py	94.2% <76.4%> (-5.8%)	:arrow_down:
pystiche_papers/ulyanov_et_al_2016/_utils.py	100.0% <100.0%> (ø)
pystiche_papers/li_wand_2016/_utils.py	96.1% <0.0%> (ø)
pystiche_papers/gatys_et_al_2017/_utils.py	100.0% <0.0%> (ø)
pystiche_papers/ulyanov_et_al_2016/_data.py	100.0% <0.0%> (ø)
pystiche_papers/ulyanov_et_al_2016/_modules.py	100.0% <0.0%> (ø)
pystiche_papers/gatys_ecker_bethge_2016/_utils.py	100.0% <0.0%> (ø)
pystiche_papers/ulyanov_et_al_2016/_nst.py	95.5% <0.0%> (+0.2%)	:arrow_up:
pystiche_papers/utils/modules.py	98.5% <0.0%> (+0.3%)	:arrow_up:

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 7ba28c6...da2503b. Read the comment docs.