Tiny fix of mmedit/apis/test.py
#261
Conversation
|
Nice. Is this a common problem for all repos? |
|
Could you pls test how this change affects the inference time? |
Codecov Report
@@ Coverage Diff @@
## master #261 +/- ##
==========================================
- Coverage 81.34% 79.93% -1.41%
==========================================
Files 158 159 +1
Lines 7773 7941 +168
Branches 1152 1177 +25
==========================================
+ Hits 6323 6348 +25
- Misses 1306 1449 +143
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.
|
mmedit/apis/test.py
Outdated
| @@ -82,6 +83,8 @@ def multi_gpu_test(model, | |||
| save_path (str): The path to save image. Default: None. | |||
| iteration (int): Iteration number. It is used for the save image name. | |||
| Default: None. | |||
| limited_gpu (bool): Limited CUDA memory or not. Default: False. | |||
There was a problem hiding this comment.
limited_gpu -> empty_cache
Empty GPU cache in each iteration or not
mmedit/apis/test.py
Outdated
| @@ -83,8 +83,7 @@ def multi_gpu_test(model, | |||
| save_path (str): The path to save image. Default: None. | |||
| iteration (int): Iteration number. It is used for the save image name. | |||
| Default: None. | |||
| limited_gpu (bool): Limited CUDA memory or not. Default: False. | |||
| If limited_gpu and not gpu_collect, empty cache in every batch. | |||
| empty_cache (bool): empty cache in every batch. Default: False. | |||
tools/test.py
Outdated
| @@ -88,7 +88,7 @@ def main(): | |||
| model = build_model(cfg.model, train_cfg=None, test_cfg=cfg.test_cfg) | |||
|
|
|||
| args.save_image = args.save_path is not None | |||
| limited_gpu = cfg.limited_gpu is not None and cfg.limited_gpu | |||
| empty_cache = cfg.empty_cache is not None and cfg.empty_cache | |||
There was a problem hiding this comment.
cfg.get('empty_cache', False)
|
I suggest adding an item in FAQ https://github.com/open-mmlab/mmediting/blob/master/docs/faq.md, to demonstrate how to use this option. |
* tiny fix * Tiny Fix, add limited_gpu. * Tiny Fix * Tiny Fix Co-authored-by: liyinshuo <[email protected]>
In the current test program (
mmedit/apis/test.py), the GPU memory occupied by the processed data is not released in time, andCUDA out of memoryif the test data is large.Fix the issue in this PR.