Skip to content

P
PaddleDetection

PaddlePaddle / PaddleDetection
大约 1 年 前同步成功

通知 695
Star 11112
Fork 2696
P
PaddleDetection
未验证 提交 ae1b479c 编写于 作者: F Feng Ni 提交者： GitHub
浏览文件
操作

comments for architectures post_process and data source (#2473)

上级 b174b027
隐藏空白更改
内联 并排
Showing with 207 addition and 85 deletion
configs/cascade_rcnn/README.md
浏览文件 @ ae1b479c
......@@ -4,8 +4,8 @@
| 骨架网络 | 网络类型 | 每张GPU图片个数 | 学习率策略 |推理时间(fps) | Box AP | Mask AP | 下载 | 配置文件 |
| :------------------- | :------------- | :-----: | :-----: | :------------: | :-----: | :-----: | :-----------------------------------------------------: | :-----: |
| ResNet50-FPN | Cascade Faster | 1 | 1x | ---- | 41.1 | - | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/cascade_rcnn/cascade_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50-FPN | Cascade Mask | 1 | 1x | ---- | 41.8 | 36.3 | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_mask_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/cascade_rcnn/cascade_mask_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50-FPN | Cascade Faster | 1 | 1x | ---- | 41.1 | - | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/cascade_rcnn/cascade_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50-FPN | Cascade Mask | 1 | 1x | ---- | 41.8 | 36.3 | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_mask_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/cascade_rcnn/cascade_mask_rcnn_r50_fpn_1x_coco.yml) |
**注意：** Cascade R-CNN模型精度依赖Paddle develop分支修改，精度复现须使用[每日版本](https://www.paddlepaddle.org.cn/documentation/docs/zh/install/Tables.html#whl-dev)或2.0.1版本(将于2021.03发布)，使用Paddle 2.0.0版本会有少量精度损失。
......
configs/dcn/README.md
浏览文件 @ ae1b479c
......@@ -2,17 +2,17 @@
| 骨架网络 | 网络类型 | 卷积 | 每张GPU图片个数 | 学习率策略 |推理时间(fps)| Box AP | Mask AP | 下载 | 配置文件 |
| :------------------- | :------------- | :-----: |:--------: | :-----: | :-----------: |:----: | :-----: | :----------------------------------------------------------: | :----: |
| ResNet50-FPN | Faster | c3-c5 | 1 | 1x | - | 42.1 | - | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_dcn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/faster_rcnn_dcn_r50_fpn_1x_coco.yml) |
| ResNet50-vd-FPN | Faster | c3-c5 | 1 | 1x | - | 42.7 | - | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_dcn_r50_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/faster_rcnn_dcn_r50_vd_fpn_1x_coco.yml) |
| ResNet50-vd-FPN | Faster | c3-c5 | 1 | 2x | - | 43.7 | - | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_dcn_r50_vd_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/faster_rcnn_dcn_r50_vd_fpn_2x_coco.yml) |
| ResNet101-vd-FPN | Faster | c3-c5 | 1 | 1x | - | 45.1 | - | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_dcn_r101_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/faster_rcnn_dcn_r101_vd_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Faster | c3-c5 | 1 | 1x | - | 46.5 | - | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.pdparams) |[配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/faster_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml) |
| ResNet50-FPN | Mask | c3-c5 | 1 | 1x | - | 42.7 | 38.4 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_dcn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/mask_rcnn_dcn_r50_fpn_1x_coco.yml) |
| ResNet50-vd-FPN | Mask | c3-c5 | 1 | 2x | - | 44.6 | 39.8 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_dcn_r50_vd_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/mask_rcnn_dcn_r50_vd_fpn_2x_coco.yml) |
| ResNet101-vd-FPN | Mask | c3-c5 | 1 | 1x | - | 45.6 | 40.6 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_dcn_r101_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/mask_rcnn_dcn_r101_vd_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Mask | c3-c5 | 1 | 1x | - | 47.3 | 42.0 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/mask_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml) |
| ResNet50-FPN | Cascade Faster | c3-c5 | 1 | 1x | - | 42.1 | - | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_rcnn_dcn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/cascade_rcnn_dcn_r50_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Cascade Faster | c3-c5 | 1 | 1x | - | 48.8 | - | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/dcn/cascade_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml) |
| ResNet50-FPN | Faster | c3-c5 | 1 | 1x | - | 42.1 | - | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_dcn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/faster_rcnn_dcn_r50_fpn_1x_coco.yml) |
| ResNet50-vd-FPN | Faster | c3-c5 | 1 | 1x | - | 42.7 | - | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_dcn_r50_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/faster_rcnn_dcn_r50_vd_fpn_1x_coco.yml) |
| ResNet50-vd-FPN | Faster | c3-c5 | 1 | 2x | - | 43.7 | - | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_dcn_r50_vd_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/faster_rcnn_dcn_r50_vd_fpn_2x_coco.yml) |
| ResNet101-vd-FPN | Faster | c3-c5 | 1 | 1x | - | 45.1 | - | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_dcn_r101_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/faster_rcnn_dcn_r101_vd_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Faster | c3-c5 | 1 | 1x | - | 46.5 | - | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.pdparams) |[配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/faster_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml) |
| ResNet50-FPN | Mask | c3-c5 | 1 | 1x | - | 42.7 | 38.4 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_dcn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/mask_rcnn_dcn_r50_fpn_1x_coco.yml) |
| ResNet50-vd-FPN | Mask | c3-c5 | 1 | 2x | - | 44.6 | 39.8 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_dcn_r50_vd_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/mask_rcnn_dcn_r50_vd_fpn_2x_coco.yml) |
| ResNet101-vd-FPN | Mask | c3-c5 | 1 | 1x | - | 45.6 | 40.6 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_dcn_r101_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/mask_rcnn_dcn_r101_vd_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Mask | c3-c5 | 1 | 1x | - | 47.3 | 42.0 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/mask_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml) |
| ResNet50-FPN | Cascade Faster | c3-c5 | 1 | 1x | - | 42.1 | - | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_rcnn_dcn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/cascade_rcnn_dcn_r50_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Cascade Faster | c3-c5 | 1 | 1x | - | 48.8 | - | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/dcn/cascade_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml) |
**注意事项:**
......
configs/faster_rcnn/README.md
浏览文件 @ ae1b479c
......@@ -4,20 +4,20 @@
| 骨架网络 | 网络类型 | 每张GPU图片个数 | 学习率策略 |推理时间(fps) | Box AP | 下载 | 配置文件 |
| :------------------- | :------------- | :-----: | :-----: | :------------: | :-----: | :-----------------------------------------------------: | :-----: |
| ResNet50 | Faster | 1 | 1x | ---- | 36.7 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r50_1x_coco.yml) |
| ResNet50-vd | Faster | 1 | 1x | ---- | 37.6 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_vd_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r50_vd_1x_coco.yml) |
| ResNet101 | Faster | 1 | 1x | ---- | 39.0 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r101_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r101_1x_coco.yml) |
| ResNet34-FPN | Faster | 1 | 1x | ---- | 37.8 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r34_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r34_fpn_1x_coco.yml) |
| ResNet34-vd-FPN | Faster | 1 | 1x | ---- | 38.5 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r34_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r34_vd_fpn_1x_coco.yml) |
| ResNet50-FPN | Faster | 1 | 1x | ---- | 38.4 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50-FPN | Faster | 1 | 2x | ---- | 40.0 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r50_fpn_2x_coco.yml) |
| ResNet50-vd-FPN | Faster | 1 | 1x | ---- | 39.5 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r50_vd_fpn_1x_coco.yml) |
| ResNet50-vd-FPN | Faster | 1 | 2x | ---- | 40.8 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_vd_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r50_vd_fpn_2x_coco.yml) |
| ResNet101-FPN | Faster | 1 | 2x | ---- | 41.4 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r101_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r101_fpn_2x_coco.yml) |
| ResNet101-vd-FPN | Faster | 1 | 1x | ---- | 42.0 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r101_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r101_vd_fpn_1x_coco.yml) |
| ResNet101-vd-FPN | Faster | 1 | 2x | ---- | 43.0 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r101_vd_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_r101_vd_fpn_2x_coco.yml) |
| ResNeXt101-vd-FPN | Faster | 1 | 1x | ---- | 43.4 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_x101_vd_64x4d_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_x101_vd_64x4d_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Faster | 1 | 2x | ---- | 44.0 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_x101_vd_64x4d_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/faster_rcnn/faster_rcnn_x101_vd_64x4d_fpn_2x_coco.yml) |
| ResNet50 | Faster | 1 | 1x | ---- | 36.7 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r50_1x_coco.yml) |
| ResNet50-vd | Faster | 1 | 1x | ---- | 37.6 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_vd_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r50_vd_1x_coco.yml) |
| ResNet101 | Faster | 1 | 1x | ---- | 39.0 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r101_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r101_1x_coco.yml) |
| ResNet34-FPN | Faster | 1 | 1x | ---- | 37.8 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r34_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r34_fpn_1x_coco.yml) |
| ResNet34-vd-FPN | Faster | 1 | 1x | ---- | 38.5 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r34_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r34_vd_fpn_1x_coco.yml) |
| ResNet50-FPN | Faster | 1 | 1x | ---- | 38.4 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50-FPN | Faster | 1 | 2x | ---- | 40.0 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r50_fpn_2x_coco.yml) |
| ResNet50-vd-FPN | Faster | 1 | 1x | ---- | 39.5 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r50_vd_fpn_1x_coco.yml) |
| ResNet50-vd-FPN | Faster | 1 | 2x | ---- | 40.8 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_vd_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r50_vd_fpn_2x_coco.yml) |
| ResNet101-FPN | Faster | 1 | 2x | ---- | 41.4 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r101_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r101_fpn_2x_coco.yml) |
| ResNet101-vd-FPN | Faster | 1 | 1x | ---- | 42.0 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r101_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r101_vd_fpn_1x_coco.yml) |
| ResNet101-vd-FPN | Faster | 1 | 2x | ---- | 43.0 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r101_vd_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_r101_vd_fpn_2x_coco.yml) |
| ResNeXt101-vd-FPN | Faster | 1 | 1x | ---- | 43.4 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_x101_vd_64x4d_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_x101_vd_64x4d_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Faster | 1 | 2x | ---- | 44.0 | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_x101_vd_64x4d_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/faster_rcnn/faster_rcnn_x101_vd_64x4d_fpn_2x_coco.yml) |
**注意：** Faster R-CNN模型精度依赖Paddle develop分支修改，精度复现须使用[每日版本](https://www.paddlepaddle.org.cn/documentation/docs/zh/install/Tables.html#whl-dev)或2.0.1版本(将于2021.03发布)，使用Paddle 2.0.0版本会有少量精度损失。
......
configs/hrnet/README.md
浏览文件 @ ae1b479c
......@@ -30,5 +30,5 @@
| Backbone | Type | Image/gpu | Lr schd | Inf time (fps) | Box AP | Mask AP | Download | Configs |
| :---------------------- | :------------- | :-------: | :-----: | :------------: | :----: | :-----: | :----------------------------------------------------------: | :-----: |
| HRNetV2p_W18 | Faster | 1 | 1x | - | 36.8 | - | [model](https://paddledet.bj.bcebos.com/models/faster_rcnn_hrnetv2p_w18_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/hrnet/faster_rcnn_hrnetv2p_w18_1x_coco.yml) |
| HRNetV2p_W18 | Faster | 1 | 2x | - | 39.0 | - | [model](https://paddledet.bj.bcebos.com/models/faster_rcnn_hrnetv2p_w18_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/hrnet/faster_rcnn_hrnetv2p_w18_2x_coco.yml) |
| HRNetV2p_W18 | Faster | 1 | 1x | - | 36.8 | - | [model](https://paddledet.bj.bcebos.com/models/faster_rcnn_hrnetv2p_w18_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/hrnet/faster_rcnn_hrnetv2p_w18_1x_coco.yml) |
| HRNetV2p_W18 | Faster | 1 | 2x | - | 39.0 | - | [model](https://paddledet.bj.bcebos.com/models/faster_rcnn_hrnetv2p_w18_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/hrnet/faster_rcnn_hrnetv2p_w18_2x_coco.yml) |
configs/mask_rcnn/README.md
浏览文件 @ ae1b479c
......@@ -4,16 +4,16 @@
| 骨架网络 | 网络类型 | 每张GPU图片个数 | 学习率策略 |推理时间(fps) | Box AP | Mask AP | 下载 | 配置文件 |
| :------------------- | :------------| :-----: | :-----: | :------------: | :-----: | :-----: | :-----------------------------------------------------: | :-----: |
| ResNet50 | Mask | 1 | 1x | ---- | 37.4 | 32.8 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/mask_rcnn/mask_rcnn_r50_1x_coco.yml) |
| ResNet50 | Mask | 1 | 2x | ---- | 39.7 | 34.5 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/mask_rcnn/mask_rcnn_r50_2x_coco.yml) |
| ResNet50-FPN | Mask | 1 | 1x | ---- | 39.2 | 35.6 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50-FPN | Mask | 1 | 2x | ---- | 40.5 | 36.7 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/mask_rcnn/mask_rcnn_r50_fpn_2x_coco.yml) |
| ResNet50-vd-FPN | Mask | 1 | 1x | ---- | 40.3 | 36.4 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/mask_rcnn/mask_rcnn_r50_vd_fpn_1x_coco.yml) |
| ResNet50-vd-FPN | Mask | 1 | 2x | ---- | 41.4 | 37.5 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_vd_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/mask_rcnn/mask_rcnn_r50_vd_fpn_2x_coco.yml) |
| ResNet101-FPN | Mask | 1 | 1x | ---- | 40.6 | 36.6 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r101_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/mask_rcnn/mask_rcnn_r101_fpn_1x_coco.yml) |
| ResNet101-vd-FPN | Mask | 1 | 1x | ---- | 42.4 | 38.1 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r101_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/mask_rcnn/mask_rcnn_r101_vd_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Mask | 1 | 1x | ---- | 44.0 | 39.5 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_x101_vd_64x4d_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/mask_rcnn/mask_rcnn_x101_vd_64x4d_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Mask | 1 | 2x | ---- | 44.6 | 39.8 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_x101_vd_64x4d_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/mask_rcnn/mask_rcnn_x101_vd_64x4d_fpn_2x_coco.yml) |
| ResNet50 | Mask | 1 | 1x | ---- | 37.4 | 32.8 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mask_rcnn/mask_rcnn_r50_1x_coco.yml) |
| ResNet50 | Mask | 1 | 2x | ---- | 39.7 | 34.5 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mask_rcnn/mask_rcnn_r50_2x_coco.yml) |
| ResNet50-FPN | Mask | 1 | 1x | ---- | 39.2 | 35.6 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50-FPN | Mask | 1 | 2x | ---- | 40.5 | 36.7 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mask_rcnn/mask_rcnn_r50_fpn_2x_coco.yml) |
| ResNet50-vd-FPN | Mask | 1 | 1x | ---- | 40.3 | 36.4 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mask_rcnn/mask_rcnn_r50_vd_fpn_1x_coco.yml) |
| ResNet50-vd-FPN | Mask | 1 | 2x | ---- | 41.4 | 37.5 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_vd_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mask_rcnn/mask_rcnn_r50_vd_fpn_2x_coco.yml) |
| ResNet101-FPN | Mask | 1 | 1x | ---- | 40.6 | 36.6 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r101_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mask_rcnn/mask_rcnn_r101_fpn_1x_coco.yml) |
| ResNet101-vd-FPN | Mask | 1 | 1x | ---- | 42.4 | 38.1 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r101_vd_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mask_rcnn/mask_rcnn_r101_vd_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Mask | 1 | 1x | ---- | 44.0 | 39.5 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_x101_vd_64x4d_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mask_rcnn/mask_rcnn_x101_vd_64x4d_fpn_1x_coco.yml) |
| ResNeXt101-vd-FPN | Mask | 1 | 2x | ---- | 44.6 | 39.8 | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_x101_vd_64x4d_fpn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mask_rcnn/mask_rcnn_x101_vd_64x4d_fpn_2x_coco.yml) |
**注意：** Mask R-CNN模型精度依赖Paddle develop分支修改，精度复现须使用[每日版本](https://www.paddlepaddle.org.cn/documentation/docs/zh/install/Tables.html#whl-dev)或2.0.1版本(将于2021.03发布)，使用Paddle 2.0.0版本会有少量精度损失。
......
ppdet/data/source/category.py
浏览文件 @ ae1b479c
......@@ -32,6 +32,8 @@ def get_categories(metric_type, anno_file=None):
to category name map from annotation file.
Args:
metric_type (str): metric type, currently support 'coco', 'voc', 'oid'
and 'widerface'.
anno_file (str): annotation file path
"""
if metric_type.lower() == 'coco':
......
ppdet/data/source/coco.py
浏览文件 @ ae1b479c
......@@ -24,6 +24,17 @@ logger = setup_logger(__name__)
@register
@serializable
class COCODataSet(DetDataset):
"""
Load dataset with COCO format.
Args:
dataset_dir (str): root directory for dataset.
image_dir (str): directory for images.
anno_path (str): coco annotation file path.
data_fields (list): key name of data dictionary, at least have 'image'.
sample_num (int): number of samples to load, -1 means all.
"""
def __init__(self,
dataset_dir=None,
image_dir=None,
......
ppdet/data/source/dataset.py
浏览文件 @ ae1b479c
......@@ -27,6 +27,18 @@ import copy
@serializable
class DetDataset(Dataset):
"""
Load detection dataset.
Args:
dataset_dir (str): root directory for dataset.
image_dir (str): directory for images.
anno_path (str): annotation file path.
data_fields (list): key name of data dictionary, at least have 'image'.
sample_num (int): number of samples to load, -1 means all.
use_default_label (bool): whether to load default label list.
"""
def __init__(self,
dataset_dir=None,
image_dir=None,
......
ppdet/data/source/voc.py
浏览文件 @ ae1b479c
......@@ -38,6 +38,7 @@ class VOCDataSet(DetDataset):
dataset_dir (str): root directory for dataset.
image_dir (str): directory for images.
anno_path (str): voc annotation file path.
data_fields (list): key name of data dictionary, at least have 'image'.
sample_num (int): number of samples to load, -1 means all.
label_list (str): if use_default_label is False, will load
mapping between category and class index.
......
ppdet/data/source/widerface.py
浏览文件 @ ae1b479c
......@@ -31,8 +31,10 @@ class WIDERFaceDataSet(DetDataset):
Args:
dataset_dir (str): root directory for dataset.
image_dir (str): directory for images.
anno_path (str): root directory for voc annotation data
sample_num (int): number of samples to load, -1 means all
anno_path (str): WiderFace annotation data.
data_fields (list): key name of data dictionary, at least have 'image'.
sample_num (int): number of samples to load, -1 means all.
with_lmk (bool): whether to load face landmark keypoint labels.
"""
def __init__(self,
......
ppdet/metrics/coco_utils.py
浏览文件 @ ae1b479c
......@@ -67,13 +67,13 @@ def cocoapi_eval(jsonfile,
classwise=False):
"""
Args:
jsonfile: Evaluation json file, eg: bbox.json, mask.json.
style: COCOeval style, can be `bbox` , `segm` and `proposal`.
coco_gt: Whether to load COCOAPI through anno_file,
jsonfile (str): Evaluation json file, eg: bbox.json, mask.json.
style (str): COCOeval style, can be `bbox` , `segm` and `proposal`.
coco_gt (str): Whether to load COCOAPI through anno_file,
eg: coco_gt = COCO(anno_file)
anno_file: COCO annotations file.
max_dets: COCO evaluation maxDets.
classwise: whether per-category AP and draw P-R Curve or not.
anno_file (str): COCO annotations file.
max_dets (tuple): COCO evaluation maxDets.
classwise (bool): Whether per-category AP and draw P-R Curve or not.
"""
assert coco_gt != None or anno_file != None
from pycocotools.coco import COCO
......@@ -142,9 +142,7 @@ def cocoapi_eval(jsonfile,
return coco_eval.stats
def json_eval_results(metric: object,
json_directory: object=None,
dataset: object=None) -> object:
def json_eval_results(metric, json_directory, dataset):
"""
cocoapi eval with already exists proposal.json, bbox.json or mask.json
"""
......
ppdet/metrics/map_utils.py
浏览文件 @ ae1b479c
......@@ -101,19 +101,20 @@ class DetectionMAP(object):
Currently support two types: 11point and integral
Args:
class_num (int): the class number.
class_num (int): The class number.
overlap_thresh (float): The threshold of overlap
ratio between prediction bounding box and
ground truth bounding box for deciding
true/false positive. Default 0.5.
map_type (str): calculation method of mean average
map_type (str): Calculation method of mean average
precision, currently support '11point' and
'integral'. Default '11point'.
is_bbox_normalized (bool): whther bounding boxes
is_bbox_normalized (bool): Whether bounding boxes
is normalized to range[0, 1]. Default False.
evaluate_difficult (bool): whether to evaluate
evaluate_difficult (bool): Whether to evaluate
difficult bounding boxes. Default False.
classwise (bool): whether per-category AP and draw
catid2name (dict): Mapping between category id and category name.
classwise (bool): Whether per-category AP and draw
P-R Curve or not.
"""
......
ppdet/modeling/architectures/cascade_rcnn.py
浏览文件 @ ae1b479c
......@@ -25,6 +25,18 @@ __all__ = ['CascadeRCNN']
@register
class CascadeRCNN(BaseArch):
"""
Cascade R-CNN network, see https://arxiv.org/abs/1712.00726
Args:
backbone (object): backbone instance
rpn_head (object): `RPNHead` instance
bbox_head (object): `BBoxHead` instance
bbox_post_process (object): `BBoxPostProcess` instance
neck (object): 'FPN' instance
mask_head (object): `MaskHead` instance
mask_post_process (object): `MaskPostProcess` instance
"""
__category__ = 'architecture'
__inject__ = [
'bbox_post_process',
......
ppdet/modeling/architectures/faster_rcnn.py
浏览文件 @ ae1b479c
......@@ -25,6 +25,16 @@ __all__ = ['FasterRCNN']
@register
class FasterRCNN(BaseArch):
"""
Faster R-CNN network, see https://arxiv.org/abs/1506.01497
Args:
backbone (object): backbone instance
rpn_head (object): `RPNHead` instance
bbox_head (object): `BBoxHead` instance
bbox_post_process (object): `BBoxPostProcess` instance
neck (object): 'FPN' instance
"""
__category__ = 'architecture'
__inject__ = ['bbox_post_process']
......@@ -34,13 +44,6 @@ class FasterRCNN(BaseArch):
bbox_head,
bbox_post_process,
neck=None):
"""
backbone (nn.Layer): backbone instance.
rpn_head (nn.Layer): generates proposals using backbone features.
bbox_head (nn.Layer): a head that performs per-region computation.
mask_head (nn.Layer): generates mask from bbox and backbone features.
"""
super(FasterRCNN, self).__init__()
self.backbone = backbone
self.neck = neck
......
ppdet/modeling/architectures/mask_rcnn.py
浏览文件 @ ae1b479c
......@@ -25,6 +25,19 @@ __all__ = ['MaskRCNN']
@register
class MaskRCNN(BaseArch):
"""
Mask R-CNN network, see https://arxiv.org/abs/1703.06870
Args:
backbone (object): backbone instance
rpn_head (object): `RPNHead` instance
bbox_head (object): `BBoxHead` instance
mask_head (object): `MaskHead` instance
bbox_post_process (object): `BBoxPostProcess` instance
mask_post_process (object): `MaskPostProcess` instance
neck (object): 'FPN' instance
"""
__category__ = 'architecture'
__inject__ = [
'bbox_post_process',
......@@ -39,12 +52,6 @@ class MaskRCNN(BaseArch):
bbox_post_process,
mask_post_process,
neck=None):
"""
backbone (nn.Layer): backbone instance.
rpn_head (nn.Layer): generates proposals using backbone features.
bbox_head (nn.Layer): a head that performs per-region computation.
mask_head (nn.Layer): generates mask from bbox and backbone features.
"""
super(MaskRCNN, self).__init__()
self.backbone = backbone
self.neck = neck
......
ppdet/modeling/backbones/hrnet.py
浏览文件 @ ae1b479c
......@@ -97,7 +97,12 @@ class ConvNormLayer(nn.Layer):
class Layer1(nn.Layer):
def __init__(self, num_channels, has_se=False, freeze_norm=True, name=None):
def __init__(self,
num_channels,
has_se=False,
norm_decay=0.,
freeze_norm=True,
name=None):
super(Layer1, self).__init__()
self.bottleneck_block_list = []
......@@ -111,6 +116,7 @@ class Layer1(nn.Layer):
has_se=has_se,
stride=1,
downsample=True if i == 0 else False,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name=name + '_' + str(i + 1)))
self.bottleneck_block_list.append(bottleneck_block)
......@@ -123,7 +129,12 @@ class Layer1(nn.Layer):
class TransitionLayer(nn.Layer):
def __init__(self, in_channels, out_channels, freeze_norm=True, name=None):
def __init__(self,
in_channels,
out_channels,
norm_decay=0.,
freeze_norm=True,
name=None):
super(TransitionLayer, self).__init__()
num_in = len(in_channels)
......@@ -140,6 +151,7 @@ class TransitionLayer(nn.Layer):
ch_in=in_channels[i],
ch_out=out_channels[i],
filter_size=3,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act='relu',
name=name + '_layer_' + str(i + 1)))
......@@ -151,6 +163,7 @@ class TransitionLayer(nn.Layer):
ch_out=out_channels[i],
filter_size=3,
stride=2,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act='relu',
name=name + '_layer_' + str(i + 1)))
......@@ -175,6 +188,7 @@ class Branches(nn.Layer):
in_channels,
out_channels,
has_se=False,
norm_decay=0.,
freeze_norm=True,
name=None):
super(Branches, self).__init__()
......@@ -190,6 +204,7 @@ class Branches(nn.Layer):
num_channels=in_ch,
num_filters=out_channels[i],
has_se=has_se,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name=name + '_branch_layer_' + str(i + 1) + '_' +
str(j + 1)))
......@@ -213,6 +228,7 @@ class BottleneckBlock(nn.Layer):
has_se,
stride=1,
downsample=False,
norm_decay=0.,
freeze_norm=True,
name=None):
super(BottleneckBlock, self).__init__()
......@@ -224,6 +240,7 @@ class BottleneckBlock(nn.Layer):
ch_in=num_channels,
ch_out=num_filters,
filter_size=1,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act="relu",
name=name + "_conv1")
......@@ -232,6 +249,7 @@ class BottleneckBlock(nn.Layer):
ch_out=num_filters,
filter_size=3,
stride=stride,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act="relu",
name=name + "_conv2")
......@@ -239,6 +257,7 @@ class BottleneckBlock(nn.Layer):
ch_in=num_filters,
ch_out=num_filters * 4,
filter_size=1,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act=None,
name=name + "_conv3")
......@@ -248,6 +267,7 @@ class BottleneckBlock(nn.Layer):
ch_in=num_channels,
ch_out=num_filters * 4,
filter_size=1,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act=None,
name=name + "_downsample")
......@@ -283,6 +303,7 @@ class BasicBlock(nn.Layer):
stride=1,
has_se=False,
downsample=False,
norm_decay=0.,
freeze_norm=True,
name=None):
super(BasicBlock, self).__init__()
......@@ -293,6 +314,7 @@ class BasicBlock(nn.Layer):
ch_in=num_channels,
ch_out=num_filters,
filter_size=3,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
stride=stride,
act="relu",
......@@ -301,6 +323,7 @@ class BasicBlock(nn.Layer):
ch_in=num_filters,
ch_out=num_filters,
filter_size=3,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
stride=1,
act=None,
......@@ -311,6 +334,7 @@ class BasicBlock(nn.Layer):
ch_in=num_channels,
ch_out=num_filters * 4,
filter_size=1,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act=None,
name=name + "_downsample")
......@@ -381,6 +405,7 @@ class Stage(nn.Layer):
num_modules,
num_filters,
has_se=False,
norm_decay=0.,
freeze_norm=True,
multi_scale_output=True,
name=None):
......@@ -396,6 +421,7 @@ class Stage(nn.Layer):
num_channels=num_channels,
num_filters=num_filters,
has_se=has_se,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
multi_scale_output=False,
name=name + '_' + str(i + 1)))
......@@ -406,6 +432,7 @@ class Stage(nn.Layer):
num_channels=num_channels,
num_filters=num_filters,
has_se=has_se,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name=name + '_' + str(i + 1)))
......@@ -424,6 +451,7 @@ class HighResolutionModule(nn.Layer):
num_filters,
has_se=False,
multi_scale_output=True,
norm_decay=0.,
freeze_norm=True,
name=None):
super(HighResolutionModule, self).__init__()
......@@ -432,6 +460,7 @@ class HighResolutionModule(nn.Layer):
in_channels=num_channels,
out_channels=num_filters,
has_se=has_se,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name=name)
......@@ -439,6 +468,7 @@ class HighResolutionModule(nn.Layer):
in_channels=num_filters,
out_channels=num_filters,
multi_scale_output=multi_scale_output,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name=name)
......@@ -453,6 +483,7 @@ class FuseLayers(nn.Layer):
in_channels,
out_channels,
multi_scale_output=True,
norm_decay=0.,
freeze_norm=True,
name=None):
super(FuseLayers, self).__init__()
......@@ -473,6 +504,7 @@ class FuseLayers(nn.Layer):
filter_size=1,
stride=1,
act=None,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name=name + '_layer_' + str(i + 1) + '_' +
str(j + 1)))
......@@ -489,6 +521,7 @@ class FuseLayers(nn.Layer):
ch_out=out_channels[i],
filter_size=3,
stride=2,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act=None,
name=name + '_layer_' + str(i + 1) + '_' +
......@@ -503,6 +536,7 @@ class FuseLayers(nn.Layer):
ch_out=out_channels[j],
filter_size=3,
stride=2,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act="relu",
name=name + '_layer_' + str(i + 1) + '_' +
......@@ -544,6 +578,7 @@ class HRNet(nn.Layer):
has_se (bool): whether to add SE block for each stage
freeze_at (int): the stage to freeze
freeze_norm (bool): whether to freeze norm in HRNet
norm_decay (float): weight decay for normalization layer weights
return_idx (List): the stage to return
"""
......@@ -586,6 +621,7 @@ class HRNet(nn.Layer):
ch_out=64,
filter_size=3,
stride=2,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act='relu',
name="layer1_1")
......@@ -595,6 +631,7 @@ class HRNet(nn.Layer):
ch_out=64,
filter_size=3,
stride=2,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
act='relu',
name="layer1_2")
......@@ -602,12 +639,14 @@ class HRNet(nn.Layer):
self.la1 = Layer1(
num_channels=64,
has_se=has_se,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name="layer2")
self.tr1 = TransitionLayer(
in_channels=[256],
out_channels=channels_2,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name="tr1")
......@@ -616,12 +655,14 @@ class HRNet(nn.Layer):
num_modules=num_modules_2,
num_filters=channels_2,
has_se=self.has_se,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name="st2")
self.tr2 = TransitionLayer(
in_channels=channels_2,
out_channels=channels_3,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name="tr2")
......@@ -630,12 +671,14 @@ class HRNet(nn.Layer):
num_modules=num_modules_3,
num_filters=channels_3,
has_se=self.has_se,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name="st3")
self.tr3 = TransitionLayer(
in_channels=channels_3,
out_channels=channels_4,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name="tr3")
self.st4 = Stage(
......@@ -643,6 +686,7 @@ class HRNet(nn.Layer):
num_modules=num_modules_4,
num_filters=channels_4,
has_se=self.has_se,
norm_decay=norm_decay,
freeze_norm=freeze_norm,
name="st4")
......
ppdet/modeling/necks/hrfpn.py
浏览文件 @ ae1b479c
......@@ -30,8 +30,8 @@ class HRFPN(nn.Layer):
in_channels (list): number of input feature channels from backbone
out_channel (int): number of output feature channels
share_conv (bool): whether to share conv for different layers' reduction
spatial_scales (list): feature map scaling factor
extra_stage (int): add extra stage for returning HRFPN fpn_feats
spatial_scales (list): feature map scaling factor
"""
def __init__(self,
......
ppdet/modeling/post_process.py
浏览文件 @ ae1b479c
......@@ -24,6 +24,8 @@ try:
except Exception:
from collections import Sequence
__all__ = ['BBoxPostProcess', 'MaskPostProcess', 'FCOSPostProcess']
@register
class BBoxPostProcess(object):
......@@ -40,13 +42,17 @@ class BBoxPostProcess(object):
"""
Decode the bbox and do NMS if needed.
Args:
head_out (tuple): bbox_pred and cls_prob of bbox_head output.
rois (tuple): roi and rois_num of rpn_head output.
im_shape (Tensor): The shape of the input image.
scale_factor (Tensor): The scale factor of the input image.
Returns:
bbox_pred(Tensor): The output is the prediction with shape [N, 6]
including labels, scores and bboxes. The size of
bboxes are corresponding to the input image and
the bboxes may be used in other brunch.
bbox_num(Tensor): The number of prediction of each batch with shape
[N, 6].
bbox_pred (Tensor): The output prediction with shape [N, 6], including
labels, scores and bboxes. The size of bboxes are corresponding
to the input image, the bboxes may be used in other branch.
bbox_num (Tensor): The number of prediction boxes of each batch with
shape [1], and is N.
"""
if self.nms is not None:
bboxes, score = self.decode(head_out, rois, im_shape, scale_factor)
......@@ -54,6 +60,9 @@ class BBoxPostProcess(object):
else:
bbox_pred, bbox_num = self.decode(head_out, rois, im_shape,
scale_factor)
# Prevent empty bbox_pred from decode or NMS.
# Bboxes and score before NMS may be empty due to the score threshold.
if bbox_pred.shape[0] == 0:
bbox_pred = paddle.to_tensor(
np.array(
......@@ -64,16 +73,22 @@ class BBoxPostProcess(object):
def get_pred(self, bboxes, bbox_num, im_shape, scale_factor):
"""
Rescale, clip and filter the bbox from the output of NMS to
get final prediction.
get final prediction.
Notes:
Currently only support bs = 1.
Args:
bboxes(Tensor): The output of __call__ with shape [N, 6]
bbox_pred (Tensor): The output bboxes with shape [N, 6] after decode
and NMS, including labels, scores and bboxes.
bbox_num (Tensor): The number of prediction boxes of each batch with
shape [1], and is N.
im_shape (Tensor): The shape of the input image.
scale_factor (Tensor): The scale factor of the input image.
Returns:
bbox_pred(Tensor): The output is the prediction with shape [N, 6]
including labels, scores and bboxes. The size of
bboxes are corresponding to the original image.
pred_result (Tensor): The final prediction results with shape [N, 6]
including labels, scores and bboxes.
"""
origin_shape = paddle.floor(im_shape / scale_factor + 0.5)
origin_shape_list = []
......@@ -125,7 +140,9 @@ class MaskPostProcess(object):
self.binary_thresh = binary_thresh
def paste_mask(self, masks, boxes, im_h, im_w):
# paste each mask on image
"""
Paste the mask prediction to the original image.
"""
x0, y0, x1, y1 = paddle.split(boxes, 4, axis=1)
masks = paddle.unsqueeze(masks, [0, 1])
img_y = paddle.arange(0, im_h, dtype='float32') + 0.5
......@@ -148,7 +165,19 @@ class MaskPostProcess(object):
def __call__(self, mask_out, bboxes, bbox_num, origin_shape):
"""
Paste the mask prediction to the original image.
Decode the mask_out and paste the mask to the origin image.
Args:
mask_out (Tensor): mask_head output with shape [N, 28, 28].
bbox_pred (Tensor): The output bboxes with shape [N, 6] after decode
and NMS, including labels, scores and bboxes.
bbox_num (Tensor): The number of prediction boxes of each batch with
shape [1], and is N.
origin_shape (Tensor): The origin shape of the input image, the tensor
shape is [N, 2], and each row is [h, w].
Returns:
pred_result (Tensor): The final prediction mask results with shape
[N, h, w] in binary mask style.
"""
num_mask = mask_out.shape[0]
origin_shape = paddle.cast(origin_shape, 'int32')
......
Markdown is supported
0% 或 .
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 或