memory efficient test #330
Conversation
…msegmentation into memory-efficient-test Conflicts: mmseg/apis/test.py mmseg/core/evaluation/metrics.py
mmseg/apis/test.py
Outdated
| temp_file_name = tempfile.NamedTemporaryFile( | ||
| suffix='.npy', delete=False).name | ||
| np.save(temp_file_name, item) |
There was a problem hiding this comment.
I suggest making this part a function. So that we could have sth like:
results.extend(np2tmp(_) for _ in result)There was a problem hiding this comment.
Is there a pair of []?
results.extend([np2tmp(_) for _ in result])
| if reduce_zero_label: | ||
| # avoid using underflow conversion | ||
| label[label == 0] = 255 | ||
| label = label - 1 | ||
| label[label == 254] = 255 |
There was a problem hiding this comment.
Could reduce_zero_labels be merged with label_map?
There was a problem hiding this comment.
But in custom.py, reduce_zero_label and label_map is two different attributes. If we merge the two different attributes in metrics.py but retain the two different attributes in custom.py, is it something disunity?
|
Codecov Report
@@ Coverage Diff @@
## master #330 +/- ##
==========================================
- Coverage 86.47% 86.07% -0.41%
==========================================
Files 95 95
Lines 4761 4789 +28
Branches 765 778 +13
==========================================
+ Hits 4117 4122 +5
- Misses 499 519 +20
- Partials 145 148 +3
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
|
CPU memory efficient test DeeplabV3+ on Cityscapes (without saving the test results) and evaluate the mIoU. Using |
* memory efficient test * implement efficient test * merge * Add document and docstring * fix unit test * add memory usage report
* add void check * remove void, add types for params
CPU memory efficient test DeeplabV3+ on Cityscapes (without saving the test results) and evaluate the mIoU.
Using
pmapto view CPU memory footprint, it used 2.25GB CPU memory withefficient_test=Trueand 11.06GB CPU memory withefficient_test=False. This optional parameter can save a lot of memory.