Update weights and re-evaluate models #274
Conversation
Codecov Report
@@ Coverage Diff @@
## master #274 +/- ##
=======================================
Coverage 79.94% 79.94%
=======================================
Files 159 159
Lines 7942 7942
Branches 1177 1177
=======================================
Hits 6349 6349
Misses 1449 1449
Partials 144 144
Flags with carried forward coverage won't be shown. Click here to find out more. Continue to review full report at Codecov.
|
| @@ -16,7 +16,7 @@ | |||
| pixel_loss=dict(type='CharbonnierLoss', loss_weight=1.0, reduction='mean')) | |||
| # model training and testing settings | |||
| train_cfg = dict(fix_iter=5000) | |||
| test_cfg = dict(metrics=['PSNR', 'SSIM'], crop_border=0, convert_to='y') | |||
There was a problem hiding this comment.
Is removing 'SSIM' intended?
There was a problem hiding this comment.
Yes, as it would slow down the training. Therefore I put only PSNR here.
There was a problem hiding this comment.
Users may remove 'SSIM', or together with 'PSNR' themselves if they want to speed up training. It's better to leave it here since our result page contains both SSIM and PSNR.
|
Invalid link, will update |
* Update weights and re-evaluate models * Update links to weights * Add back SSIM to configuration
When
channel_orderis not used correctly, BasicVSR and IconVSR output BGR image hence the evaluation result is not correct (since the functionbgr2grayis used when evaluating in Y-channel). This PR re-computes the results in the datasets and updates the links for the pre-trained weights.