提交 e1d60414 编写于 作者: F FlyingQianMM

add r18_fpn pretrained weights for rcnn

上级 0180098e
...@@ -36,6 +36,7 @@ ...@@ -36,6 +36,7 @@
| 模型 | 模型大小 | 预测时间(毫秒) | BoxAP(%) | | 模型 | 模型大小 | 预测时间(毫秒) | BoxAP(%) |
|:-------|:-----------|:-------------|:----------| |:-------|:-----------|:-------------|:----------|
|[FasterRCNN-ResNet18-FPN](https://bj.bcebos.com/paddlex/pretrained_weights/faster_rcnn_r18_fpn_1x.tar) | 173.2M | - | 32.6 |
|[FasterRCNN-ResNet50](https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r50_1x.tar)|136.0MB| 197.715 | 35.2 | |[FasterRCNN-ResNet50](https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r50_1x.tar)|136.0MB| 197.715 | 35.2 |
|[FasterRCNN-ResNet50_vd](https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r50_vd_1x.tar)| 136.1MB | 475.700 | 36.4 | |[FasterRCNN-ResNet50_vd](https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r50_vd_1x.tar)| 136.1MB | 475.700 | 36.4 |
|[FasterRCNN-ResNet101](https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r101_1x.tar)| 212.5MB | 582.911 | 38.3 | |[FasterRCNN-ResNet101](https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r101_1x.tar)| 212.5MB | 582.911 | 38.3 |
...@@ -55,6 +56,7 @@ ...@@ -55,6 +56,7 @@
| 模型 | 模型大小 | 预测时间(毫秒) | BoxAP (%) | MaskAP (%) | | 模型 | 模型大小 | 预测时间(毫秒) | BoxAP (%) | MaskAP (%) |
|:-------|:-----------|:-------------|:----------|:----------| |:-------|:-----------|:-------------|:----------|:----------|
|[MaskRCNN-ResNet18-FPN](https://bj.bcebos.com/paddlex/pretrained_weights/mask_rcnn_r18_fpn_1x.tar) | 189.1MB | - | 33.6 | 30.5 |
|[MaskRCNN-ResNet50](https://paddlemodels.bj.bcebos.com/object_detection/mask_rcnn_r50_2x.tar) | 143.9MB | 87 | 38.2 | 33.4 | |[MaskRCNN-ResNet50](https://paddlemodels.bj.bcebos.com/object_detection/mask_rcnn_r50_2x.tar) | 143.9MB | 87 | 38.2 | 33.4 |
|[MaskRCNN-ResNet50-FPN](https://paddlemodels.bj.bcebos.com/object_detection/mask_rcnn_r50_fpn_2x.tar)| 177.7MB | 63.9 | 38.7 | 34.7 | |[MaskRCNN-ResNet50-FPN](https://paddlemodels.bj.bcebos.com/object_detection/mask_rcnn_r50_fpn_2x.tar)| 177.7MB | 63.9 | 38.7 | 34.7 |
|[MaskRCNN-ResNet50_vd-FPN](https://paddlemodels.bj.bcebos.com/object_detection/mask_rcnn_r50_vd_fpn_2x.tar) | 177.7MB | 63.1 | 39.8 || 35.4 | |[MaskRCNN-ResNet50_vd-FPN](https://paddlemodels.bj.bcebos.com/object_detection/mask_rcnn_r50_vd_fpn_2x.tar) | 177.7MB | 63.1 | 39.8 || 35.4 |
......
...@@ -10,9 +10,9 @@ PaddleX目前提供了MaskRCNN实例分割模型结构,多种backbone模型, ...@@ -10,9 +10,9 @@ PaddleX目前提供了MaskRCNN实例分割模型结构,多种backbone模型,
| 模型(点击获取代码) | Box MMAP/Seg MMAP | 模型大小 | GPU预测速度 | Arm预测速度 | 备注 | | 模型(点击获取代码) | Box MMAP/Seg MMAP | 模型大小 | GPU预测速度 | Arm预测速度 | 备注 |
| :---------------- | :------- | :------- | :--------- | :--------- | :----- | | :---------------- | :------- | :------- | :--------- | :--------- | :----- |
| [MaskRCNN-ResNet50-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/instance_segmentation/mask_rcnn_r50_fpn.py) | 38.7%/34.7% | 170.0MB | 160.185ms | - | 模型精度高,适用于服务端部署 | | [MaskRCNN-ResNet50-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/instance_segmentation/mask_rcnn_r50_fpn.py) | 38.7%/34.7% | 177.7MB | 160.185ms | - | 模型精度高,适用于服务端部署 |
| [MaskRCNN-ResNet18-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/instance_segmentation/mask_rcnn_r18_fpn.py) | -/- | 120.0MB | - | - | 模型精度高,适用于服务端部署 | | [MaskRCNN-ResNet18-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/instance_segmentation/mask_rcnn_r18_fpn.py) | 33.6/30.5 | 189.1MB | - | - | 模型精度高,适用于服务端部署 |
| [MaskRCNN-HRNet-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/instance_segmentation/mask_rcnn_hrnet_fpn.py) | 38.7%/34.7% | 116.MB | - | - | 模型精度高,预测速度快,适用于服务端部署 | | [MaskRCNN-HRNet-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/instance_segmentation/mask_rcnn_hrnet_fpn.py) | 38.7%/34.7% | 120.7MB | - | - | 模型精度高,预测速度快,适用于服务端部署 |
## 开始训练 ## 开始训练
......
...@@ -13,8 +13,8 @@ PaddleX目前提供了FasterRCNN和YOLOv3两种检测结构,多种backbone模型 ...@@ -13,8 +13,8 @@ PaddleX目前提供了FasterRCNN和YOLOv3两种检测结构,多种backbone模型
| [YOLOv3-MobileNetV1](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/yolov3_mobilenetv1.py) | 29.3% | 99.2MB | 15.442ms | - | 模型小,预测速度快,适用于低性能或移动端设备 | | [YOLOv3-MobileNetV1](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/yolov3_mobilenetv1.py) | 29.3% | 99.2MB | 15.442ms | - | 模型小,预测速度快,适用于低性能或移动端设备 |
| [YOLOv3-MobileNetV3](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/yolov3_mobilenetv3.py) | 31.6% | 100.7MB | 143.322ms | - | 模型小,移动端上预测速度有优势 | | [YOLOv3-MobileNetV3](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/yolov3_mobilenetv3.py) | 31.6% | 100.7MB | 143.322ms | - | 模型小,移动端上预测速度有优势 |
| [YOLOv3-DarkNet53](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/yolov3_darknet53.py) | 38.9 | 249.2MB | 42.672ms | - | 模型较大,预测速度快,适用于服务端 | | [YOLOv3-DarkNet53](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/yolov3_darknet53.py) | 38.9 | 249.2MB | 42.672ms | - | 模型较大,预测速度快,适用于服务端 |
| [FasterRCNN-ResNet50-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/faster_rcnn_r50_fpn.py) | 37.2% | 136.0MB | 197.715ms | - | 模型精度高,适用于服务端部署 | | [FasterRCNN-ResNet50-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/faster_rcnn_r50_fpn.py) | 37.2% | 167.7MB | 197.715ms | - | 模型精度高,适用于服务端部署 |
| [FasterRCNN-ResNet18-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/faster_rcnn_r18_fpn.py) | - | - | - | - | 模型精度高,适用于服务端部署 | | [FasterRCNN-ResNet18-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/faster_rcnn_r18_fpn.py) | 32.6% | 173.2MB | - | - | 模型精度高,适用于服务端部署 |
| [FasterRCNN-HRNet-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/faster_rcnn_hrnet_fpn.py) | 36.0% | 115.MB | 81.592ms | - | 模型精度高,预测速度快,适用于服务端部署 | | [FasterRCNN-HRNet-FPN](https://github.com/PaddlePaddle/PaddleX/blob/develop/tutorials/train/object_detection/faster_rcnn_hrnet_fpn.py) | 36.0% | 115.MB | 81.592ms | - | 模型精度高,预测速度快,适用于服务端部署 |
......
...@@ -88,6 +88,8 @@ coco_pretrain = { ...@@ -88,6 +88,8 @@ coco_pretrain = {
'https://paddlemodels.bj.bcebos.com/object_detection/yolov3_r34.tar', 'https://paddlemodels.bj.bcebos.com/object_detection/yolov3_r34.tar',
'YOLOv3_ResNet50_vd_COCO': 'YOLOv3_ResNet50_vd_COCO':
'https://paddlemodels.bj.bcebos.com/object_detection/yolov3_r50vd_dcn.tar', 'https://paddlemodels.bj.bcebos.com/object_detection/yolov3_r50vd_dcn.tar',
'FasterRCNN_ResNet18_COCO':
'https://bj.bcebos.com/paddlex/pretrained_weights/faster_rcnn_r18_fpn_1x.tar',
'FasterRCNN_ResNet50_COCO': 'FasterRCNN_ResNet50_COCO':
'https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r50_fpn_2x.tar', 'https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r50_fpn_2x.tar',
'FasterRCNN_ResNet50_vd_COCO': 'FasterRCNN_ResNet50_vd_COCO':
...@@ -98,6 +100,8 @@ coco_pretrain = { ...@@ -98,6 +100,8 @@ coco_pretrain = {
'https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r101_vd_fpn_2x.tar', 'https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r101_vd_fpn_2x.tar',
'FasterRCNN_HRNet_W18_COCO': 'FasterRCNN_HRNet_W18_COCO':
'https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_hrnetv2p_w18_2x.tar', 'https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_hrnetv2p_w18_2x.tar',
'MaskRCNN_ResNet18_COCO':
'https://bj.bcebos.com/paddlex/pretrained_weights/mask_rcnn_r18_fpn_1x.tar',
'MaskRCNN_ResNet50_COCO': 'MaskRCNN_ResNet50_COCO':
'https://paddlemodels.bj.bcebos.com/object_detection/mask_rcnn_r50_fpn_2x.tar', 'https://paddlemodels.bj.bcebos.com/object_detection/mask_rcnn_r50_fpn_2x.tar',
'MaskRCNN_ResNet50_vd_COCO': 'MaskRCNN_ResNet50_vd_COCO':
...@@ -136,9 +140,10 @@ def get_pretrain_weights(flag, class_name, backbone, save_dir): ...@@ -136,9 +140,10 @@ def get_pretrain_weights(flag, class_name, backbone, save_dir):
return flag return flag
warning_info = "{} does not support to be finetuned with weights pretrained on the {} dataset, so pretrain_weights is forced to be set to {}" warning_info = "{} does not support to be finetuned with weights pretrained on the {} dataset, so pretrain_weights is forced to be set to {}"
if flag == 'COCO': if flag == 'COCO':
if class_name == "FasterRCNN" and backbone in ['ResNet18'] or \ if class_name == 'DeepLabv3p' and backbone in [
class_name == "MaskRCNN" and backbone in ['ResNet18'] or \ 'Xception41', 'MobileNetV2_x0.25', 'MobileNetV2_x0.5',
class_name == 'DeepLabv3p' and backbone in ['Xception41', 'MobileNetV2_x0.25', 'MobileNetV2_x0.5', 'MobileNetV2_x1.5', 'MobileNetV2_x2.0']: 'MobileNetV2_x1.5', 'MobileNetV2_x2.0'
]:
model_name = '{}_{}'.format(class_name, backbone) model_name = '{}_{}'.format(class_name, backbone)
logging.warning(warning_info.format(model_name, flag, 'IMAGENET')) logging.warning(warning_info.format(model_name, flag, 'IMAGENET'))
flag = 'IMAGENET' flag = 'IMAGENET'
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册