Merge branch 'master' of github.com:open-mmlab/mmgeneration into pt19_gpu_fuser

LeoXing1996 · LeoXing1996 · commit c261dcdabf2c · 2022-01-07T17:50:55.000+08:00
diff --git a/README.md b/README.md
@@ -156,3 +156,6 @@ If you find this project useful in your research, please consider cite:
 - [MMFlow](https://github.com/open-mmlab/mmflow): OpenMMLab optical flow toolbox and benchmark.
 - [MMFewShot](https://github.com/open-mmlab/mmfewshot): OpenMMLab FewShot Learning Toolbox and Benchmark.
 - [MMHuman3D](https://github.com/open-mmlab/mmhuman3d): OpenMMLab Human Pose and Shape Estimation Toolbox and Benchmark.
+- [MMSelfSup](https://github.com/open-mmlab/mmselfsup): OpenMMLab self-supervised learning toolbox and benchmark.
+- [MMRazor](https://github.com/open-mmlab/mmrazor): OpenMMLab Model Compression Toolbox and Benchmark.
+- [MMDeploy](https://github.com/open-mmlab/mmdeploy): OpenMMLab Model Deployment Framework.
diff --git a/README_zh-CN.md b/README_zh-CN.md
@@ -158,6 +158,9 @@ v0.3.0 在 02/08/2021 发布。 关于细节和发布历史，请参考 [changel
 - [MMFlow](https://github.com/open-mmlab/mmflow): OpenMMLab 光流估计工具箱与测试基准
 - [MMFewShot](https://github.com/open-mmlab/mmfewshot): OpenMMLab 少样本学习工具箱与测试基准
 - [MMHuman3D](https://github.com/open-mmlab/mmhuman3d): OpenMMLab 人体参数化模型工具箱与测试基准
+- [MMSelfSup](https://github.com/open-mmlab/mmselfsup): OpenMMLab 自监督学习工具箱与测试基准
+- [MMRazor](https://github.com/open-mmlab/mmrazor): OpenMMLab 模型压缩工具箱与测试基准
+- [MMDeploy](https://github.com/open-mmlab/mmdeploy): OpenMMLab 模型部署框架
 
 ## 欢迎加入 OpenMMLab 社区
 
diff --git a/configs/positional_encoding_in_gans/singan_csg_bohemian.py b/configs/positional_encoding_in_gans/singan_csg_bohemian.py
@@ -1,4 +1,4 @@
-_base_ = ['../singan/singan_fish.py']
+_base_ = ['../singan/singan_bohemian.py']
 
 num_scales = 10  # start from zero
 model = dict(
@@ -21,5 +21,5 @@
         max_size=500,
     ))
 
-dist_params = dict(backend='nccl', port=28120)
+dist_params = dict(backend='nccl')
 total_iters = 22000
diff --git a/configs/positional_encoding_in_gans/singan_csg_fish.py b/configs/positional_encoding_in_gans/singan_csg_fish.py
@@ -16,10 +16,10 @@
 
 data = dict(
     train=dict(
-        img_path='./data/singan/fish.jpg',
+        img_path='./data/singan/fish-crop.jpg',
         min_size=25,
         max_size=300,
     ))
 
-dist_params = dict(backend='nccl', port=28113)
+dist_params = dict(backend='nccl')
 total_iters = 22000
diff --git a/configs/positional_encoding_in_gans/singan_interp-pad_balloons.py b/configs/positional_encoding_in_gans/singan_interp-pad_balloons.py
@@ -7,4 +7,4 @@
 
 train_cfg = dict(fixed_noise_with_pad=True)
 
-dist_params = dict(backend='nccl', port=23019)
+dist_params = dict(backend='nccl')
diff --git a/configs/positional_encoding_in_gans/singan_interp-pad_disc-nobn_balloons.py b/configs/positional_encoding_in_gans/singan_interp-pad_disc-nobn_balloons.py
@@ -8,4 +8,4 @@
 
 train_cfg = dict(fixed_noise_with_pad=True)
 
-dist_params = dict(backend='nccl', port=23119)
+dist_params = dict(backend='nccl')
diff --git a/configs/positional_encoding_in_gans/singan_interp-pad_disc-nobn_fish.py b/configs/positional_encoding_in_gans/singan_interp-pad_disc-nobn_fish.py
@@ -8,4 +8,11 @@
 
 train_cfg = dict(fixed_noise_with_pad=True)
 
-dist_params = dict(backend='nccl', port=23119)
+data = dict(
+    train=dict(
+        img_path='./data/singan/fish-crop.jpg',
+        min_size=25,
+        max_size=300,
+    ))
+
+dist_params = dict(backend='nccl')
diff --git a/configs/positional_encoding_in_gans/singan_spe-dim4_bohemian.py b/configs/positional_encoding_in_gans/singan_spe-dim4_bohemian.py
@@ -23,10 +23,10 @@
 
 data = dict(
     train=dict(
-        img_path='./data/singan/fish.jpg',
+        img_path='./data/singan/bohemian.png',
         min_size=25,
-        max_size=300,
+        max_size=500,
     ))
 
-dist_params = dict(backend='nccl', port=28119)
+dist_params = dict(backend='nccl')
 total_iters = 22000
diff --git a/configs/positional_encoding_in_gans/singan_spe-dim4_fish.py b/configs/positional_encoding_in_gans/singan_spe-dim4_fish.py
@@ -21,10 +21,10 @@
 
 data = dict(
     train=dict(
-        img_path='./data/singan/fish.jpg',
+        img_path='./data/singan/fish-crop.jpg',
         min_size=25,
         max_size=300,
     ))
 
-dist_params = dict(backend='nccl', port=28119)
+dist_params = dict(backend='nccl')
 total_iters = 22000
diff --git a/configs/positional_encoding_in_gans/singan_spe-dim8_bohemian.py b/configs/positional_encoding_in_gans/singan_spe-dim8_bohemian.py
@@ -23,10 +23,10 @@
 
 data = dict(
     train=dict(
-        img_path='./data/singan/fish.jpg',
+        img_path='./data/singan/bohemian.png',
         min_size=25,
-        max_size=300,
+        max_size=500,
     ))
 
-dist_params = dict(backend='nccl', port=28119)
+dist_params = dict(backend='nccl')
 total_iters = 22000
diff --git a/configs/singan/singan_fish.py b/configs/singan/singan_fish.py
@@ -19,7 +19,8 @@
 # )
 
 data = dict(
-    train=dict(img_path='./data/singan/fish.png', min_size=25, max_size=300))
+    train=dict(
+        img_path='./data/singan/fish-crop.jpg', min_size=25, max_size=300))
 
 optimizer = None
 lr_config = None
diff --git a/docker/Dockerfile b/docker/Dockerfile
@@ -0,0 +1,24 @@
+ARG PYTORCH="1.8.0"
+ARG CUDA="11.1"
+ARG CUDNN="8"
+
+FROM pytorch/pytorch:${PYTORCH}-cuda${CUDA}-cudnn${CUDNN}-devel
+
+ENV TORCH_CUDA_ARCH_LIST="6.0 6.1 7.0+PTX"
+ENV TORCH_NVCC_FLAGS="-Xfatbin -compress-all"
+ENV CMAKE_PREFIX_PATH="$(dirname $(which conda))/../"
+
+RUN apt-get update && apt-get install -y ffmpeg libsm6 libxext6 git ninja-build libglib2.0-0 libsm6 libxrender-dev libxext6 \
+    && apt-get clean \
+    && rm -rf /var/lib/apt/lists/*
+
+# Install MMCV
+RUN pip install mmcv-full==1.3.16 -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.8.0/index.html
+
+# Install MMFlow
+RUN conda clean --all
+RUN git clone https://github.com/open-mmlab/mmgeneration.git /mmgen
+WORKDIR /mmgen
+ENV FORCE_CUDA="1"
+RUN pip install -r requirements.txt
+RUN pip install --no-cache-dir -e .
diff --git a/mmgen/datasets/pipelines/augmentation.py b/mmgen/datasets/pipelines/augmentation.py
@@ -149,23 +149,32 @@ def __init__(self,
         self.interpolation = interpolation
         self.backend = backend
 
-    def _resize(self, img):
+    def _resize(self, img, scale):
+        """Resize given image with corresponding scale.
+        Args:
+            img (np.array): Image to be resized.
+            scale (float | Tuple[int]): Scale used in resize process.
+
+        Returns:
+            tuple: Tuple contains resized image and scale factor in resize
+                process.
+        """
         if self.keep_ratio:
-            img, self.scale_factor = mmcv.imrescale(
+            img, scale_factor = mmcv.imrescale(
                 img,
-                self.scale,
+                scale,
                 return_scale=True,
                 interpolation=self.interpolation,
                 backend=self.backend)
         else:
             img, w_scale, h_scale = mmcv.imresize(
                 img,
-                self.scale,
+                scale,
                 return_scale=True,
                 interpolation=self.interpolation,
                 backend=self.backend)
-            self.scale_factor = np.array((w_scale, h_scale), dtype=np.float32)
-        return img
+            scale_factor = np.array((w_scale, h_scale), dtype=np.float32)
+        return img, scale_factor
 
     def __call__(self, results):
         """Call function.
@@ -186,13 +195,25 @@ def __call__(self, results):
                             new_h)
                 new_w = min(self.max_size - (self.max_size % self.size_factor),
                             new_w)
-            self.scale = (new_w, new_h)
+            scale = (new_w, new_h)
+        elif isinstance(self.scale, tuple) and (np.inf in self.scale):
+            # find inf in self.scale, calculate ``scale`` manually
+            h, w = results[self.keys[0]].shape[:2]
+            if h < w:
+                scale = (int(self.scale[-1] / h * w), self.scale[-1])
+            else:
+                scale = (self.scale[-1], int(self.scale[-1] / w * h))
+        else:
+            # direct use the given ones
+            scale = self.scale
+
+        # here we assume all images in self.keys have the same input size
         for key in self.keys:
-            results[key] = self._resize(results[key])
+            results[key], scale_factor = self._resize(results[key], scale)
             if len(results[key].shape) == 2:
                 results[key] = np.expand_dims(results[key], axis=2)
 
-        results['scale_factor'] = self.scale_factor
+        results['scale_factor'] = scale_factor
         results['keep_ratio'] = self.keep_ratio
         results['interpolation'] = self.interpolation
 
diff --git a/mmgen/datasets/singan_dataset.py b/mmgen/datasets/singan_dataset.py
@@ -23,15 +23,15 @@ def create_real_pyramid(real, min_size, max_size, scale_factor_init):
 
     num_scales = int(
         np.ceil(
-            np.log(
-                np.power(min_size / min(real.shape[0], real.shape[1]), 1),
-                scale_factor_init))) + 1
+            np.log(np.power(min_size / min(real.shape[0], real.shape[1]), 1)) /
+            np.log(scale_factor_init))) + 1
 
     scale2stop = int(
         np.ceil(
             np.log(
                 min([max_size, max([real.shape[0], real.shape[1]])]) /
-                max([real.shape[0], real.shape[1]]), scale_factor_init)))
+                max([real.shape[0], real.shape[1]])) /
+            np.log(scale_factor_init)))
 
     stop_scale = num_scales - scale2stop
 
@@ -45,7 +45,8 @@ def create_real_pyramid(real, min_size, max_size, scale_factor_init):
         np.ceil(
             np.log(
                 min([max_size, max([real.shape[0], real.shape[1]])]) /
-                max([real.shape[0], real.shape[1]]), scale_factor_init)))
+                max([real.shape[0], real.shape[1]])) /
+            np.log(scale_factor_init)))
     stop_scale = num_scales - scale2stop
 
     reals = []
diff --git a/requirements/docs.txt b/requirements/docs.txt
@@ -1,7 +1,7 @@
 click
 docutils==0.16.0
 m2r
-mmcls==0.10.0
+mmcls==0.18.0
 myst-parser
 opencv-python
 prettytable
diff --git a/requirements/mminstall.txt b/requirements/mminstall.txt
@@ -1,2 +1,2 @@
-mmcls>=0.10.0,<=0.10.0
+mmcls>=0.18.0,<=0.18.0
 mmcv-full>=1.3.0,<=1.4.0
diff --git a/requirements/runtime.txt b/requirements/runtime.txt
@@ -1,4 +1,4 @@
-mmcls==0.10.0
+mmcls==0.18.0
 numpy
 prettytable
 scikit-image
diff --git a/tests/test_datasets/test_pipelines/test_augmentation.py b/tests/test_datasets/test_pipelines/test_augmentation.py
@@ -174,7 +174,16 @@ def test_resize(self):
         assert resize_results['alpha'].shape == (192, 192, 1)
 
         resize = Resize(['gt_img'], (-1, 200))
+        results = dict(gt_img=self.results['gt'].copy())
+        resize_results = resize(results)
+        assert resize.scale == (np.inf, 200)
+        assert resize_results['gt_img'].shape == (400, 200, 3)
+
+        resize = Resize(['gt_img'], (-1, 200))
+        results = dict(gt_img=self.results['gt'].copy().transpose(1, 0, 2))
+        resize_results = resize(results)
         assert resize.scale == (np.inf, 200)
+        assert resize_results['gt_img'].shape == (200, 400, 3)
 
         results = dict(gt_img=self.results['img'].copy())
         resize_keep_ratio = Resize(['gt_img'], scale=0.5, keep_ratio=True)
diff --git a/tests/test_losses/test_pixelwise_loss.py b/tests/test_losses/test_pixelwise_loss.py
@@ -182,7 +182,7 @@ def test_disc_gaussian_log_likelihood_loss(self):
         disc_gaussian_loss = DiscretizedGaussianLogLikelihoodLoss(
             loss_weight=weight, data_info=self.data_info, reduction='mean')
         loss = disc_gaussian_loss(self.output_dict)
-        assert torch.allclose(loss, weight.mean() * self.gt_loss)
+        assert torch.allclose(loss, weight.mean() * self.gt_loss, atol=1e-6)
 
         # test weight --> tensor & batchmean
         weight = torch.randn(*self.tar_shape)
@@ -191,5 +191,5 @@ def test_disc_gaussian_log_likelihood_loss(self):
             data_info=self.data_info,
             reduction='batchmean')
         loss = disc_gaussian_loss(self.output_dict)
-        assert torch.allclose(loss,
-                              weight.sum([1, 2, 3]).mean() * self.gt_loss)
+        assert torch.allclose(
+            loss, weight.sum([1, 2, 3]).mean() * self.gt_loss, atol=1e-6)
diff --git a/tests/test_modules/test_stylev1_archs.py b/tests/test_modules/test_stylev1_archs.py
@@ -179,6 +179,13 @@ def test_g_cpu(self):
             n_target=4)
         assert style_mixing_images.shape == (25, 3, 32, 32)
 
+        # set mix_prob as 1.0 to force cover lines
+        cfg_ = deepcopy(self.default_cfg)
+        cfg_['mix_prob'] = 1
+        g = StyleGANv1Generator(**cfg_)
+        res = g(None, num_batches=2)
+        assert res.shape == (2, 3, 256, 256)
+
 
 class TestStyleGANv1Disc:
 
diff --git a/tests/test_modules/test_stylev2_archs.py b/tests/test_modules/test_stylev2_archs.py
@@ -177,6 +177,13 @@ def test_stylegan2_g_cpu(self):
         res = g(None, num_batches=2)
         assert res.shape == (2, 3, 256, 256)
 
+        # set mix_prob as 1.0 to force cover lines
+        cfg_ = deepcopy(self.default_cfg)
+        cfg_['mix_prob'] = 1
+        g = StyleGANv2Generator(**cfg_)
+        res = g(None, num_batches=2)
+        assert res.shape == (2, 3, 64, 64)
+
     @pytest.mark.skipif(not torch.cuda.is_available(), reason='requires cuda')
     def test_fp16_stylegan2_G_cuda(self):
 

Original file line number	Diff line number	Diff line change
`@@ -7,4 +7,4 @@`
`7`	`7`
`8`	`8`	`train_cfg = dict(fixed_noise_with_pad=True)`
`9`	`9`
`10`		`-dist_params = dict(backend='nccl', port=23019)`
	`10`	`+dist_params = dict(backend='nccl')`
Original file line number	Diff line number	Diff line change
`@@ -8,4 +8,4 @@`
`8`	`8`
`9`	`9`	`train_cfg = dict(fixed_noise_with_pad=True)`
`10`	`10`
`11`		`-dist_params = dict(backend='nccl', port=23119)`
	`11`	`+dist_params = dict(backend='nccl')`