feat(segmentation): migrate to functional api 1.0 (#69)

Author: Jianfeng Wang

README.md

@@ -72,7 +72,7 @@ export PYTHONPATH=/path/to/models:$PYTHONPATH
 
 ### 目标检测
 
-目标检测同样是计算机视觉中的常见任务，我们提供了两个经典的目标检测模型[Retinanet](./official/vision/detection/models/retinanet.py)和[Faster R-CNN](./official/vision/detection/models/faster_rcnn.py)，这两个模型在**COCO验证集**上的测试结果如下：
+目标检测同样是计算机视觉中的常见任务，我们提供了多个经典的目标检测模型，这些模型在COCO2017验证集上的测试结果如下：
 
 | 模型                                | mAP<br>@5-95 |
 | ---                                 | :---:        |
@@ -81,21 +81,27 @@ export PYTHONPATH=/path/to/models:$PYTHONPATH
 | retinanet-resx101-coco-2x-800size   | 42.7         |
 | faster-rcnn-res50-coco-1x-800size   | 38.0         |
 | faster-rcnn-res101-coco-2x-800size  | 42.5         |
-| faster-rcnn-resx101-coco-2x-800size | 44.7 *       |
+| faster-rcnn-resx101-coco-2x-800size | 43.6         |
 | fcos-res50-coco-1x-800size          | 39.7         |
 | fcos-res101-coco-2x-800size         | 44.1         |
-| fcos-resx101-coco-2x-800size        | 39.7 *       |
+| fcos-resx101-coco-2x-800size        | 44.9         |
 | atss-res50-coco-1x-800size          | 40.1         |
 | atss-res101-coco-2x-800size         | 44.5         |
 | atss-resx101-coco-2x-800size        | 45.9         |
 
 ### 图像分割
 
-我们也提供了经典的语义分割模型--[Deeplabv3plus](./official/vision/segmentation/)，这个模型在**PASCAL VOC验证集**上的测试结果如下：
+我们也提供了经典的语义分割模型--[DeepLabV3+](./official/vision/segmentation/)，这个模型在Pascal VOC2012验证集上的测试结果如下：
 
- | 模型          | Backbone  | mIoU_single | mIoU_multi |
- | --            | :--:      | :--:        | :--:       |
- | Deeplabv3plus | Resnet101 | 79.0        | 79.8       |
+| 模型                             | mIoU |
+| ---                              | :--: |
+| deeplabv3plus-res101-voc-512size | 79.5 |
+
+在Cityscapes验证集上的测试结果如下：
+
+| 模型                                    | mIoU |
+| ---                                     | :--: |
+| deeplabv3plus-res101-cityscapes-768size | 78.5 |
 
 ### 人体关节点检测
 

hubconf.py

@@ -29,15 +29,19 @@
 )
 from official.vision.detection.configs import (
     faster_rcnn_res50_coco_1x_800size,
-    faster_rcnn_res50_coco_1x_800size_syncbn,
     faster_rcnn_res101_coco_2x_800size,
     faster_rcnn_resx101_coco_2x_800size,
     retinanet_res50_coco_1x_800size,
-    retinanet_res50_coco_1x_800size_syncbn,
     retinanet_res101_coco_2x_800size,
     retinanet_resx101_coco_2x_800size,
+    fcos_res50_coco_1x_800size,
+    fcos_res101_coco_2x_800size,
+    fcos_resx101_coco_2x_800size,
+    atss_res50_coco_1x_800size,
+    atss_res101_coco_2x_800size,
+    atss_resx101_coco_2x_800size,
 )
-from official.vision.detection.models import FasterRCNN, RetinaNet
+from official.vision.detection.models import FasterRCNN, RetinaNet, FCOS, ATSS
 from official.vision.detection.tools.utils import DetEvaluator
 from official.vision.keypoints.inference import KeypointEvaluator
 from official.vision.keypoints.models import (
@@ -46,7 +50,8 @@
     simplebaseline_res101,
     simplebaseline_res152,
 )
-from official.vision.segmentation.deeplabv3plus import (
-    DeepLabV3Plus,
-    deeplabv3plus_res101,
+from official.vision.segmentation.configs import (
+    deeplabv3plus_res101_cityscapes_768size,
+    deeplabv3plus_res101_voc_512size,
 )
+from official.vision.segmentation.models import DeepLabV3Plus

official/assets/cat_seg_out.jpg

@@ -2,7 +2,12 @@
 
 ## 介绍
 
-本目录包含了采用MegEngine实现的经典网络结构，包括[RetinaNet](https://arxiv.org/pdf/1708.02002>)、[Faster R-CNN](https://arxiv.org/pdf/1612.03144.pdf)等，同时提供了在COCO2017数据集上的完整训练和测试代码。
+本目录包含了采用MegEngine实现的如下经典网络结构，并提供了在COCO2017数据集上的完整训练和测试代码：
+
+- [RetinaNet](https://arxiv.org/abs/1708.02002)
+- [Faster R-CNN](https://arxiv.org/abs/1612.03144)
+- [FCOS](https://arxiv.org/abs/1904.01355)
+- [ATSS](https://arxiv.org/abs/1912.02424)
 
 网络在COCO2017验证集上的性能和结果如下：
 
@@ -13,10 +18,10 @@
 | retinanet-resx101-coco-2x-800size   | 42.7         | 2             |
 | faster-rcnn-res50-coco-1x-800size   | 38.0         | 2             |
 | faster-rcnn-res101-coco-2x-800size  | 42.5         | 2             |
-| faster-rcnn-resx101-coco-2x-800size | 44.7 *       | 2             |
+| faster-rcnn-resx101-coco-2x-800size | 43.6         | 2             |
 | fcos-res50-coco-1x-800size          | 39.7         | 2             |
 | fcos-res101-coco-2x-800size         | 44.1         | 2             |
-| fcos-resx101-coco-2x-800size        | 39.7 *       | 2             |
+| fcos-resx101-coco-2x-800size        | 44.9         | 2             |
 | atss-res50-coco-1x-800size          | 40.1         | 2             |
 | atss-res101-coco-2x-800size         | 44.5         | 2             |
 | atss-resx101-coco-2x-800size        | 45.9         | 2             |
@@ -119,7 +124,7 @@ python3 tools/test.py -f configs/retinanet_res50_coco_1x_800size.py -n 8 \
 
 ## 参考文献
 
-- [Focal Loss for Dense Object Detection](https://arxiv.org/pdf/1708.02002) Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, Piotr Dollár. Proceedings of the IEEE international conference on computer vision. 2017: 2980-2988.
-- [Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks](https://arxiv.org/pdf/1506.01497.pdf) S. Ren, K. He, R. Girshick, and J. Sun. In: Neural Information Processing Systems(NIPS)(2015).
-- [Feature Pyramid Networks for Object Detection](https://arxiv.org/pdf/1612.03144.pdf) T. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan and S. Belongie. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, 2017, pp. 936-944, doi: 10.1109/CVPR.2017.106.
-- [Microsoft COCO: Common Objects in Context](https://arxiv.org/pdf/1405.0312.pdf)  Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Dollár, Piotr and Zitnick, C Lawrence, Lin T Y, Maire M, Belongie S, et al. European conference on computer vision. Springer, Cham, 2014: 740-755.
+- [Microsoft COCO: Common Objects in Context](https://arxiv.org/abs/1405.0312) Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Dollár, Piotr and Zitnick, C Lawrence, Lin T Y, Maire M, Belongie S, et al. European conference on computer vision. Springer, Cham, 2014: 740-755.
+- [Focal Loss for Dense Object Detection](https://arxiv.org/abs/1708.02002) Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, Piotr Dollár. Proceedings of the IEEE international conference on computer vision. 2017: 2980-2988.
+- [Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks](https://arxiv.org/abs/1506.01497) S. Ren, K. He, R. Girshick, and J. Sun. In: Neural Information Processing Systems(NIPS)(2015).
+- [Feature Pyramid Networks for Object Detection](https://arxiv.org/abs/1612.03144) T. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan and S. Belongie. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, 2017, pp. 936-944, doi: 10.1109/CVPR.2017.106.

official/vision/detection/README.md

@@ -1,11 +1,15 @@
 from .faster_rcnn_res50_coco_1x_800size import faster_rcnn_res50_coco_1x_800size
-from .faster_rcnn_res50_coco_1x_800size_syncbn import faster_rcnn_res50_coco_1x_800size_syncbn
 from .faster_rcnn_res101_coco_2x_800size import faster_rcnn_res101_coco_2x_800size
 from .faster_rcnn_resx101_coco_2x_800size import faster_rcnn_resx101_coco_2x_800size
 from .retinanet_res50_coco_1x_800size import retinanet_res50_coco_1x_800size
-from .retinanet_res50_coco_1x_800size_syncbn import retinanet_res50_coco_1x_800size_syncbn
 from .retinanet_res101_coco_2x_800size import retinanet_res101_coco_2x_800size
 from .retinanet_resx101_coco_2x_800size import retinanet_resx101_coco_2x_800size
+from .fcos_res50_coco_1x_800size import fcos_res50_coco_1x_800size
+from .fcos_res101_coco_2x_800size import fcos_res101_coco_2x_800size
+from .fcos_resx101_coco_2x_800size import fcos_resx101_coco_2x_800size
+from .atss_res50_coco_1x_800size import atss_res50_coco_1x_800size
+from .atss_res101_coco_2x_800size import atss_res101_coco_2x_800size
+from .atss_resx101_coco_2x_800size import atss_resx101_coco_2x_800size
 
 _EXCLUDE = {}
 __all__ = [k for k in globals().keys() if k not in _EXCLUDE and not k.startswith("_")]

official/vision/detection/configs/__init__.py

@@ -24,7 +24,7 @@ def __init__(self):
 
 @hub.pretrained(
     "https://data.megengine.org.cn/models/weights/"
-    "faster_rcnn_resx101_coco_2x_800size_44dot7_d03b05b2.pkl"
+    "faster_rcnn_resx101_coco_2x_800size_43dot6_79fb71a7.pkl"
 )
 def faster_rcnn_resx101_coco_2x_800size(**kwargs):
     r"""

official/vision/detection/configs/faster_rcnn_res50_objects365_1x_800size.py

@@ -24,7 +24,7 @@ def __init__(self):
 
 @hub.pretrained(
     "https://data.megengine.org.cn/models/weights/"
-    "fcos_resx101_coco_2x_800size_39dot7_313ef718.pkl"
+    "fcos_resx101_coco_2x_800size_44dot9_37e7b921.pkl"
 )
 def fcos_resx101_coco_2x_800size(**kwargs):
     r"""

official/vision/detection/configs/faster_rcnn_resx101_coco_2x_800size.py

@@ -40,6 +40,7 @@ def main():
     cfg.backbone_pretrained = False
     model = current_network.Net(cfg)
     model.eval()
+
     state_dict = mge.load(args.weight_file)
     if "state_dict" in state_dict:
         state_dict = state_dict["state_dict"]