[Feature] Support TAM #595
Conversation
Codecov Report
@@ Coverage Diff @@
## master #595 +/- ##
==========================================
+ Coverage 84.26% 84.43% +0.16%
==========================================
Files 124 126 +2
Lines 8771 8872 +101
Branches 1467 1481 +14
==========================================
+ Hits 7391 7491 +100
+ Misses 1022 1020 -2
- Partials 358 361 +3
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
|
This model is almost ready. Please review it when you are free. |
|
Cool! @dreamerlin if you have time, please take a first pass on the code :) |
configs/recognition/tanet/tanet_r50_dense_1x1x8_100e_kinetics400_rgb.py
Outdated
Show resolved
Hide resolved
| type='SGD', | ||
| constructor='TSMOptimizerConstructor', | ||
| paramwise_cfg=dict(fc_lr5=True), | ||
| lr=0.02, # this lr is used for 8 gpus |
There was a problem hiding this comment.
In most cases, we use 64 (8 gpus x 8 batchs) samples for lr=0.01. please check it.
There was a problem hiding this comment.
TANet performs better with setting lr=0.02. I'm testing it on sthv1.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
This comment has been minimized.
This comment has been minimized.
|
kindly ping :) |
| type='SGD', | ||
| constructor='TSMOptimizerConstructor', | ||
| paramwise_cfg=dict(fc_lr5=True), | ||
| lr=0.02, # this lr is used for 8 gpus |
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
configs/recognition/tanet/tanet_r50_dense_1x1x8_100e_kinetics400_rgb.py
Outdated
Show resolved
Hide resolved
|
@innerlee Maybe @dreamerlin has done it. Could you take a moment to review the code? |
|
Okay, soon |
This comment has been minimized.
This comment has been minimized.
Support TAM