[Feature]: Add PixMIM config

configs/selfsup/pixmim/pixmim_vit-base-p16_8xb512-amp-coslr-300e_in1k.py

@@ -0,0 +1,27 @@
+_base_ = '../mae/mae_vit-base-p16_8xb512-amp-coslr-300e_in1k.py'
+
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='mmcls.ToPIL', to_rgb=True),
+    dict(type='mmcls.torchvision/Resize', size=224),
+    dict(
+        type='mmcls.torchvision/RandomCrop',
+        size=224,
+        padding=4,
+        padding_mode='reflect'),
+    dict(type='mmcls.torchvision/RandomHorizontalFlip', p=0.5),
+    dict(type='mmcls.ToNumpy', to_rgb=True),
+    dict(type='PackSelfSupInputs', meta_keys=['img_path'])
+]
+
+train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
+
+# model settings
+model = dict(
+    type='PixMIM',
+    target_generator=dict(
+        type='LowFreqTargetGenerator', radius=40, img_size=224),
+)
+
+# randomness
+randomness = dict(seed=2, diff_rank_seed=True)

mmselfsup/models/algorithms/pixmim.py

@@ -22,6 +22,7 @@ class PixMIM(MAE):
              data_samples: List[SelfSupDataSample],
              **kwargs) -> Dict[str, torch.Tensor]:
         """The forward function in training.
+
         Args:
             inputs (List[torch.Tensor]): The input images.
             data_samples (List[SelfSupDataSample]): All elements required