mirror of
https://github.com/open-mmlab/mmsegmentation.git
synced 2025-06-03 22:03:48 +08:00
35ff78a07f
Thanks for your contribution and we appreciate it a lot. The following
instructions would make your pull request more healthy and more easily
get feedback. If you do not understand some items, don't worry, just
make the pull request and seek help from maintainers.
## Motivation
Please describe the motivation of this PR and the goal you want to
achieve through this PR.
Support metrics for the depth estimation task, including RMSE, ABSRel,
and etc.
## Modification
Please briefly describe what modification is made in this PR.
## BC-breaking (Optional)
Does the modification introduce changes that break the
backward-compatibility of the downstream repos?
If so, please describe how it breaks the compatibility and how the
downstream projects should modify their code to keep compatibility with
this PR.
## Use cases (Optional)
Using the following configuration to compute depth metrics on NYU
```python
dataset_type = 'NYUDataset'
data_root = 'data/nyu'
test_pipeline = [
dict(type='LoadImageFromFile'),
dict(dict(type='LoadDepthAnnotation', depth_rescale_factor=1e-3)),
dict(
type='PackSegInputs',
meta_keys=('img_path', 'depth_map_path', 'ori_shape', 'img_shape',
'pad_shape', 'scale_factor', 'flip', 'flip_direction',
'category_id'))
]
val_dataloader = dict(
batch_size=1,
num_workers=4,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
test_mode=True,
data_prefix=dict(
img_path='images/test', depth_map_path='annotations/test'),
pipeline=test_pipeline))
test_dataloader = val_dataloader
val_evaluator = dict(type='DepthMetric', max_depth_eval=10.0, crop_type='nyu')
test_evaluator = val_evaluator
```
Example log:
## Checklist
1. Pre-commit or other linting tools are used to fix the potential lint
issues.
2. The modification is covered by complete unit tests. If not, please
add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects,
this PR should be tested with downstream projects, like MMDet or
MMDet3D.
4. The documentation has been modified accordingly, like docstring or
example tutorials.
86 lines
2.8 KiB
Python
86 lines
2.8 KiB
Python
# Copyright (c) OpenMMLab. All rights reserved.
|
|
import os.path as osp
|
|
import shutil
|
|
from unittest import TestCase
|
|
|
|
import torch
|
|
from mmengine.structures import PixelData
|
|
|
|
from mmseg.evaluation import DepthMetric
|
|
from mmseg.structures import SegDataSample
|
|
|
|
|
|
class TestDepthMetric(TestCase):
|
|
|
|
def _demo_mm_inputs(self,
|
|
batch_size=2,
|
|
image_shapes=(3, 64, 64),
|
|
num_classes=5):
|
|
"""Create a superset of inputs needed to run test or train batches.
|
|
|
|
Args:
|
|
batch_size (int): batch size. Default to 2.
|
|
image_shapes (List[tuple], Optional): image shape.
|
|
Default to (3, 64, 64)
|
|
num_classes (int): number of different classes.
|
|
Default to 5.
|
|
"""
|
|
if isinstance(image_shapes, list):
|
|
assert len(image_shapes) == batch_size
|
|
else:
|
|
image_shapes = [image_shapes] * batch_size
|
|
|
|
data_samples = []
|
|
for idx in range(batch_size):
|
|
image_shape = image_shapes[idx]
|
|
_, h, w = image_shape
|
|
|
|
data_sample = SegDataSample()
|
|
gt_depth_map = torch.rand((1, h, w)) * 10
|
|
data_sample.gt_depth_map = PixelData(data=gt_depth_map)
|
|
|
|
data_samples.append(data_sample.to_dict())
|
|
|
|
return data_samples
|
|
|
|
def _demo_mm_model_output(self,
|
|
data_samples,
|
|
batch_size=2,
|
|
image_shapes=(3, 64, 64),
|
|
num_classes=5):
|
|
|
|
_, h, w = image_shapes
|
|
|
|
for data_sample in data_samples:
|
|
data_sample['pred_depth_map'] = dict(data=torch.randn(1, h, w))
|
|
|
|
data_sample[
|
|
'img_path'] = 'tests/data/pseudo_dataset/imgs/00000_img.jpg'
|
|
return data_samples
|
|
|
|
def test_evaluate(self):
|
|
"""Test using the metric in the same way as Evalutor."""
|
|
|
|
data_samples = self._demo_mm_inputs()
|
|
data_samples = self._demo_mm_model_output(data_samples)
|
|
|
|
depth_metric = DepthMetric()
|
|
depth_metric.process([0] * len(data_samples), data_samples)
|
|
res = depth_metric.compute_metrics(depth_metric.results)
|
|
self.assertIsInstance(res, dict)
|
|
|
|
# test save depth map file in output_dir
|
|
depth_metric = DepthMetric(output_dir='tmp')
|
|
depth_metric.process([0] * len(data_samples), data_samples)
|
|
assert osp.exists('tmp')
|
|
assert osp.isfile('tmp/00000_img.png')
|
|
shutil.rmtree('tmp')
|
|
|
|
# test format_only
|
|
depth_metric = DepthMetric(output_dir='tmp', format_only=True)
|
|
depth_metric.process([0] * len(data_samples), data_samples)
|
|
assert depth_metric.results == []
|
|
assert osp.exists('tmp')
|
|
assert osp.isfile('tmp/00000_img.png')
|
|
shutil.rmtree('tmp')
|