[Refactor] Support training siamrpn++ based on BaseSOTDataset instead of SOTTrainDataset #496
Conversation
… into refactor_siamrpn
Codecov Report
@@ Coverage Diff @@
## master #496 +/- ##
==========================================
- Coverage 73.58% 73.22% -0.37%
==========================================
Files 126 126
Lines 7370 7424 +54
Branches 1380 1393 +13
==========================================
+ Hits 5423 5436 +13
- Misses 1526 1560 +34
- Partials 421 428 +7
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
|
The training iterations in this code are about 3 times larger than that of original code in every epoch. Werid. |
|
This PR can be reviewed and merged. The docs about newly released models will be updated in the next PR. |
| @@ -10,6 +10,11 @@ | |||
|
|
|||
| data_root = 'data/' | |||
| train_pipeline = [ | |||
There was a problem hiding this comment.
The train_pipeleine can be inherit from base config?
There was a problem hiding this comment.
The train_pipeline of OTB adds a SeqGrayAug, which can't inherit from the base config.
| dict(type='SeqDefaultFormatBundle', ref_prefix='search') | ||
| ] | ||
| # dataset settings | ||
| data = dict( | ||
| samples_per_gpu=16, | ||
| train=[ | ||
| train=dict(dataset_cfgs=[ |
There was a problem hiding this comment.
The key of train in data can be inherit from base config?
There was a problem hiding this comment.
Because the train_pipeline is different, the dataset_cfgs list can not inherit from the base conifg.
No description provided.