Support ConvNeXt (#93)

* add ConvNext models * add register * record bugs * fix convnext bug * add convnext base small large * add more convnext models * add all convnext models * update test result and docs
Oneflow-Inc · Feb 1, 2022 · 50e0043 · 50e0043
1 parent af25c41
commit 50e0043
Show file tree

Hide file tree

Showing 6 changed files with 848 additions and 2 deletions.
diff --git a/docs/source/changelog.md b/docs/source/changelog.md
@@ -1,13 +1,14 @@
 ## Changelog
 
-### v0.1.0 (xx/01/2022)
+### v0.1.0 (10/02/2022)
 
 **New Features**
 
 - Support `trunc_normal_` in `flowvision.layers.weight_init` [#92](https://github.com/Oneflow-Inc/vision/pull/92)
 - Support [DeiT](https://arxiv.org/abs/2012.12877) model [#115](https://github.com/Oneflow-Inc/vision/pull/115)
 - Support `PolyLRScheduler` and `TanhLRScheduler` in `flowvision.scheduler` [#85](https://github.com/Oneflow-Inc/vision/pull/85)
 - Add `resmlp_12_224_dino` model and pretrained weight [#128](https://github.com/Oneflow-Inc/vision/pull/128)
+- Support [ConvNeXt](https://arxiv.org/abs/2201.03545) model [#93](https://github.com/Oneflow-Inc/vision/pull/93)
 
 **Bug Fixes**
 
@@ -30,6 +31,7 @@
 - Add `Getting Started` docs [#124](https://github.com/Oneflow-Inc/vision/pull/124)
 - Add `resmlp_12_224_dino` docs [#128](https://github.com/Oneflow-Inc/vision/pull/128)
 - Fix `VGG` docs bug [#128](https://github.com/Oneflow-Inc/vision/pull/128)
+- Add `ConvNeXt` docs [#93](https://github.com/Oneflow-Inc/vision/pull/93)
 
 
 **Contributors**

diff --git a/docs/source/flowvision.models.rst b/docs/source/flowvision.models.rst
@@ -35,6 +35,7 @@ architectures for image classification:
 -  `ResMLP`_
 -  `gMLP`_
 -  `ConvMixer`_
+-  `ConvNeXt`_
 
 
 .. _AlexNet: https://arxiv.org/abs/1404.5997
@@ -64,6 +65,7 @@ architectures for image classification:
 .. _gMLP: https://arxiv.org/abs/2105.08050
 .. _ConvMixer: https://openreview.net/pdf?id=TVHS5Y4dNvM
 .. _EfficientNet: https://arxiv.org/abs/1905.11946
+.. _ConvNeXt: https://arxiv.org/abs/2201.03545
 
 .. currentmodule:: flowvision.models
 
@@ -365,6 +367,29 @@ ConvMixer
         convmixer_1024_20,
 
 
+ConvNeXt
+--------
+.. automodule:: flowvision.models
+    :members:
+        convnext_tiny_224,
+        convnext_small_224,
+        convnext_base_224,
+        convnext_base_384,
+        convnext_large_224,
+        convnext_large_384,
+        convnext_base_224_22k,
+        convnext_base_224_22k_to_1k,
+        convnext_base_384_22k_to_1k,
+        convnext_large_224_22k,
+        convnext_large_224_22k_to_1k,
+        convnext_large_384_22k_to_1k,
+        convnext_xlarge_224_22k,
+        convnext_xlarge_224_22k_to_1k,
+        convnext_xlarge_384_22k_to_1k,
+        convnext_iso_small_224,
+        convnext_iso_base_224,
+        convnext_iso_large_224,
+
 
 Neural Style
 ============

diff --git a/flowvision/models/__init__.py b/flowvision/models/__init__.py
@@ -22,6 +22,7 @@
 from .res2net import *
 from .efficientnet import *
 from .vision_transformer import *
+from .convnext import *
 
 from . import neural_style_transfer
 from . import detection