未验证 提交 124a8a68 编写于 作者: G Guanghua Yu 提交者: GitHub

refine dygraph config (#2051)

* refine dygraph config, test=dygraph

* fix fcos config
上级 49b6a0ec
_BASE_: [
'./_base_/models/cascade_mask_rcnn_r50_fpn.yml',
'./_base_/optimizers/rcnn_1x.yml',
'./_base_/datasets/coco_instance.yml',
'./_base_/readers/mask_fpn_reader.yml',
'./_base_/runtime.yml',
]
# Cascade R-CNN: High Quality Object Detection and Instance Segmentation
## Model Zoo
| 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://paddlemodels.bj.bcebos.com/object_detection/dygraph/cascade_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/cascade_faster_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50-FPN | Cascade Mask | 1 | 1x | ---- | 41.6 | 35.3 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/cascade_mask_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/cascade_mask_rcnn_r50_fpn_1x_coco.yml) |
## Citations
```
@article{Cai_2019,
title={Cascade R-CNN: High Quality Object Detection and Instance Segmentation},
ISSN={1939-3539},
url={http://dx.doi.org/10.1109/tpami.2019.2956516},
DOI={10.1109/tpami.2019.2956516},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
publisher={Institute of Electrical and Electronics Engineers (IEEE)},
author={Cai, Zhaowei and Vasconcelos, Nuno},
year={2019},
pages={1–1}
}
```
architecture: CascadeRCNN architecture: CascadeRCNN
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar
weights: output/cascade_mask_rcnn_r50_fpn_1x_coco/model_final
load_static_weights: True load_static_weights: True
roi_stages: 3 roi_stages: 3
......
architecture: CascadeRCNN architecture: CascadeRCNN
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar
weights: output/cascade_rcnn_r50_fpn_1x_coco/model_final
load_static_weights: True load_static_weights: True
roi_stages: 3 roi_stages: 3
......
_BASE_: [
'../datasets/coco_instance.yml',
'../runtime.yml',
'_base_/optimizer_1x.yml',
'_base_/cascade_mask_rcnn_r50_fpn.yml',
'_base_/cascade_mask_fpn_reader.yml',
]
weights: output/cascade_mask_rcnn_r50_fpn_1x_coco/model_final
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_1x.yml',
'_base_/cascade_rcnn_r50_fpn.yml',
'_base_/cascade_fpn_reader.yml',
]
weights: output/cascade_rcnn_r50_fpn_1x_coco/model_final
_BASE_: [
'./_base_/models/cascade_rcnn_r50_fpn.yml',
'./_base_/optimizers/rcnn_1x.yml',
'./_base_/datasets/coco_detection.yml',
'./_base_/readers/faster_fpn_reader.yml',
'./_base_/runtime.yml',
]
# Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
## Model Zoo
| 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 | Faster | 1 | 1x | ---- | 35.1 | - | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/faster_rcnn_r50_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/faster_rcnn_r50_1x_coco.yml) |
| ResNet50-FPN | Faster | 1 | 1x | ---- | 37.0 | - | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/faster_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/faster_rcnn_r50_fpn_1x_coco.yml) |
## Citations
```
@article{Ren_2017,
title={Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
publisher={Institute of Electrical and Electronics Engineers (IEEE)},
author={Ren, Shaoqing and He, Kaiming and Girshick, Ross and Sun, Jian},
year={2017},
month={Jun},
}
```
worker_num: 2
TrainReader:
sample_transforms:
- DecodeOp: { }
- RandomFlipImage: {prob: 0.5}
- NormalizeImage: {is_channel_first: false, is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}
- ResizeImage: {target_size: 800, max_size: 1333, interp: 1, use_cv2: true}
- Permute: {to_bgr: false, channel_first: true}
batch_transforms:
- PadBatch: {pad_to_stride: 32, use_padded_im_info: false, pad_gt: true}
batch_size: 1
shuffle: true
drop_last: true
EvalReader:
sample_transforms:
- DecodeOp: { }
- NormalizeImageOp: { is_scale: true, mean: [ 0.485,0.456,0.406 ], std: [ 0.229, 0.224,0.225 ] }
- ResizeOp: { interp: 1, target_size: [ 800, 1333 ], keep_ratio: True }
- PermuteOp: { }
batch_transforms:
- PadBatchOp: { pad_to_stride: 32, pad_gt: false }
batch_size: 1
shuffle: false
drop_last: false
drop_empty: false
TestReader:
sample_transforms:
- DecodeOp: { }
- NormalizeImageOp: { is_scale: true, mean: [ 0.485,0.456,0.406 ], std: [ 0.229, 0.224,0.225 ] }
- ResizeOp: { interp: 1, target_size: [ 800, 1333 ], keep_ratio: True }
- PermuteOp: { }
batch_transforms:
- PadBatchOp: { pad_to_stride: 32, pad_gt: false }
batch_size: 1
shuffle: false
drop_last: false
architecture: FasterRCNN architecture: FasterRCNN
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar
weights: output/faster_rcnn_r50_1x_coco/model_final.pdparams
load_static_weights: True load_static_weights: True
# Model Achitecture # Model Achitecture
......
architecture: FasterRCNN architecture: FasterRCNN
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar
weights: output/faster_rcnn_r50_fpn_1x_coco/model_final.pdparams
load_static_weights: True load_static_weights: True
# Model Achitecture # Model Achitecture
......
epoch: 12
LearningRate:
base_lr: 0.01
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones: [8, 11]
- !LinearWarmup
start_factor: 0.3333333333333333
steps: 500
OptimizerBuilder:
optimizer:
momentum: 0.9
type: Momentum
regularizer:
factor: 0.0001
type: L2
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_1x.yml',
'_base_/faster_rcnn_r50.yml',
'_base_/faster_reader.yml',
]
weights: output/faster_rcnn_r50_1x_coco/model_final
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_1x.yml',
'_base_/faster_rcnn_r50_fpn.yml',
'_base_/faster_fpn_reader.yml',
]
weights: output/faster_rcnn_r50_fpn_1x_coco/model_final
_BASE_: [
'./_base_/models/faster_rcnn_r50.yml',
'./_base_/optimizers/rcnn_1x.yml',
'./_base_/datasets/coco_detection.yml',
'./_base_/readers/faster_reader.yml',
'./_base_/runtime.yml',
]
_BASE_: [
'./_base_/models/faster_rcnn_r50_fpn.yml',
'./_base_/optimizers/rcnn_1x.yml',
'./_base_/datasets/coco_detection.yml',
'./_base_/readers/faster_fpn_reader.yml',
'./_base_/runtime.yml',
]
_BASE_: [ _BASE_: [
'../_base_/datasets/coco_detection.yml', '../datasets/coco_detection.yml',
'../_base_/runtime.yml', '../runtime.yml',
'_base_/fcos_r50_fpn.yml', '_base_/fcos_r50_fpn.yml',
'_base_/optimizer_1x.yml', '_base_/optimizer_1x.yml',
'_base_/fcos_reader.yml', '_base_/fcos_reader.yml',
......
# Mask R-CNN
## Model Zoo
| 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 | ---- | 36.4 | 31.9 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/mask_rcnn_r50_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/mask_rcnn_r50_1x_coco.yml) |
| ResNet50-FPN | Mask | 1 | 1x | ---- | 38.3 | 34.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/mask_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/mask_rcnn_r50_fpn_1x_coco.yml) |
## Citations
```
@article{He_2017,
title={Mask R-CNN},
journal={2017 IEEE International Conference on Computer Vision (ICCV)},
publisher={IEEE},
author={He, Kaiming and Gkioxari, Georgia and Dollar, Piotr and Girshick, Ross},
year={2017},
month={Oct}
}
```
worker_num: 2
TrainReader:
sample_transforms:
- DecodeOp: {}
- RandomFlipImage: {prob: 0.5, is_mask_flip: true}
- NormalizeImage: {is_channel_first: false, is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}
- ResizeImage: {target_size: 800, max_size: 1333, interp: 1, use_cv2: true}
- Permute: {to_bgr: false, channel_first: true}
batch_transforms:
- PadBatch: {pad_to_stride: 32, use_padded_im_info: false, pad_gt: true}
batch_size: 1
shuffle: true
drop_last: true
EvalReader:
sample_transforms:
- DecodeOp: {}
- NormalizeImageOp: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}
- ResizeOp: {interp: 1, target_size: [800, 1333], keep_ratio: True}
- PermuteOp: {}
batch_transforms:
- PadBatchOp: {pad_to_stride: 32, pad_gt: false}
batch_size: 1
shuffle: false
drop_last: false
drop_empty: false
TestReader:
sample_transforms:
- DecodeOp: {}
- NormalizeImageOp: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}
- ResizeOp: {interp: 1, target_size: [800, 1333], keep_ratio: True}
- PermuteOp: {}
batch_transforms:
- PadBatchOp: {pad_to_stride: 32, pad_gt: false}
batch_size: 1
shuffle: false
drop_last: false
architecture: MaskRCNN architecture: MaskRCNN
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar
weights: output/mask_rcnn_r50_fpn_1x/model_final
load_static_weights: True load_static_weights: True
# Model Achitecture # Model Achitecture
......
architecture: MaskRCNN architecture: MaskRCNN
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar
weights: output/mask_rcnn_r50_fpn_1x/model_final
load_static_weights: True load_static_weights: True
# Model Achitecture # Model Achitecture
......
epoch: 12
LearningRate:
base_lr: 0.01
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones: [8, 11]
- !LinearWarmup
start_factor: 0.3333333333333333
steps: 500
OptimizerBuilder:
optimizer:
momentum: 0.9
type: Momentum
regularizer:
factor: 0.0001
type: L2
_BASE_: [
'../datasets/coco_instance.yml',
'../runtime.yml',
'_base_/optimizer_1x.yml',
'_base_/mask_rcnn_r50.yml',
'_base_/mask_reader.yml',
]
weights: output/mask_rcnn_r50_1x_coco/model_final
_BASE_: [
'../datasets/coco_instance.yml',
'../runtime.yml',
'_base_/optimizer_1x.yml',
'_base_/mask_rcnn_r50_fpn.yml',
'_base_/mask_fpn_reader.yml',
]
weights: output/mask_rcnn_r50_fpn_1x_coco/model_final
_BASE_: [
'./_base_/models/mask_rcnn_r50.yml',
'./_base_/optimizers/rcnn_1x.yml',
'./_base_/datasets/coco_instance.yml',
'./_base_/readers/mask_reader.yml',
'./_base_/runtime.yml',
]
_BASE_: [
'./_base_/models/mask_rcnn_r50_fpn.yml',
'./_base_/optimizers/rcnn_1x.yml',
'./_base_/datasets/coco_instance.yml',
'./_base_/readers/mask_fpn_reader.yml',
'./_base_/runtime.yml',
]
# SSD: Single Shot MultiBox Detector
## Model Zoo
### SSD on Pascal VOC
| 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 |
| :-------------- | :------------- | :-----: | :-----: | :------------: | :-----: | :-----------------------------------------------------: | :-----: |
| VGG | SSD | 8 | 240e | ---- | 78.2 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/ssd_vgg16_300_240e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/ssd_vgg16_300_240e_voc.yml) |
**注意:** SSD使用4GPU训练,训练240个epoch
## Citations
```
@article{Liu_2016,
title={SSD: Single Shot MultiBox Detector},
journal={ECCV},
author={Liu, Wei and Anguelov, Dragomir and Erhan, Dumitru and Szegedy, Christian and Reed, Scott and Fu, Cheng-Yang and Berg, Alexander C.},
year={2016},
}
```
architecture: SSD architecture: SSD
pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/ssd_mobilenet_v1_coco_pretrained.tar pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/ssd_mobilenet_v1_coco_pretrained.tar
weights: output/ssd_mobilenetv1/model_final load_static_weights: True
SSD: SSD:
backbone: MobileNet backbone: MobileNet
......
architecture: SSD architecture: SSD
pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/dygraph/VGG16_caffe_pretrained.pdparams pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/dygraph/VGG16_caffe_pretrained.pdparams
weights: output/ssd_vgg16/model_final load_static_weights: True
# Model Achitecture # Model Achitecture
SSD: SSD:
......
architecture: SSD architecture: SSD
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_ssld_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_ssld_pretrained.tar
weights: output/ssdlite_mobilenet_v1/model_final load_static_weights: True
SSD: SSD:
backbone: MobileNet backbone: MobileNet
......
architecture: SSD architecture: SSD
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_large_x1_0_ssld_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_large_x1_0_ssld_pretrained.tar
weights: output/ssdlite_mobilenet_v3_large/model_final load_static_weights: True
SSD: SSD:
backbone: MobileNetV3 backbone: MobileNetV3
......
architecture: SSD architecture: SSD
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_small_x1_0_ssld_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_small_x1_0_ssld_pretrained.tar
weights: output/ssd_mobilenet_v3_small/model_final load_static_weights: True
SSD: SSD:
backbone: MobileNetV3 backbone: MobileNetV3
......
_BASE_: [
'../datasets/voc.yml',
'../runtime.yml',
'_base_/optimizer_120e.yml',
'_base_/ssd_mobilenet_v1_300.yml',
'_base_/ssd_mobilenet_reader.yml',
]
weights: output/ssd_mobilenet_v1_300_120e_voc/model_final
_BASE_: [
'../datasets/voc.yml',
'../runtime.yml',
'_base_/optimizer_240e.yml',
'_base_/ssd_vgg16_300.yml',
'_base_/ssd_reader.yml',
]
weights: output/ssd_vgg16_300_240e_voc/model_final
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_1000e.yml',
'_base_/ssdlite_mobilenet_v1_300.yml',
'_base_/ssdlite300_reader.yml',
]
weights: output/ssdlite_mobilenet_v1_300_coco/model_final
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_1000e.yml',
'_base_/ssdlite_mobilenet_v3_large_320.yml',
'_base_/ssdlite320_reader.yml',
]
weights: output/ssdlite_mobilenet_v3_large_320_coco/model_final
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_1000e.yml',
'_base_/ssdlite_mobilenet_v3_small_320.yml',
'_base_/ssdlite320_reader.yml',
]
weights: output/ssdlite_mobilenet_v3_small_320_coco/model_final
_BASE_: [
'./_base_/models/ssd_mobilenet_v1_300.yml',
'./_base_/optimizers/ssd_mobilenet_120e.yml',
'./_base_/datasets/voc.yml',
'./_base_/readers/ssd_mobilenet_reader.yml',
'./_base_/runtime.yml',
]
_BASE_: [
'./_base_/models/ssd_vgg16_300.yml',
'./_base_/optimizers/ssd_240e.yml',
'./_base_/datasets/voc.yml',
'./_base_/readers/ssd_reader.yml',
'./_base_/runtime.yml',
]
_BASE_: [
'./_base_/models/ssdlite_mobilenet_v1_300.yml',
'./_base_/optimizers/ssdlite_1000e.yml',
'./_base_/datasets/coco_detection.yml',
'./_base_/readers/ssdlite300_reader.yml',
'./_base_/runtime.yml',
]
_BASE_: [
'./_base_/models/ssdlite_mobilenet_v3_large_320.yml',
'./_base_/optimizers/ssdlite_1000e.yml',
'./_base_/datasets/coco_detection.yml',
'./_base_/readers/ssdlite320_reader.yml',
'./_base_/runtime.yml',
]
_BASE_: [
'./_base_/models/ssdlite_mobilenet_v3_small_320.yml',
'./_base_/optimizers/ssdlite_1000e.yml',
'./_base_/datasets/coco_detection.yml',
'./_base_/readers/ssdlite320_reader.yml',
'./_base_/runtime.yml',
]
# YOLOv3
## Model Zoo
### YOLOv3 on COCO
| 骨架网络 | 输入尺寸 | 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 |
| :------------------- | :------- | :-----: | :-----: | :------------: | :-----: | :-----------------------------------------------------: | :-----: |
| DarkNet53(paper) | 608 | 8 | 270e | ---- | 33.0 | - | - |
| DarkNet53(paper) | 416 | 8 | 270e | ---- | 31.0 | - | - |
| DarkNet53(paper) | 320 | 8 | 270e | ---- | 28.2 | - | - |
| DarkNet53 | 608 | 8 | 270e | ---- | 39.0 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_darknet53_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_darknet53_270e_coco.yml) |
| DarkNet53 | 416 | 8 | 270e | ---- | 37.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_darknet53_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_darknet53_270e_coco.yml) |
| DarkNet53 | 320 | 8 | 270e | ---- | 34.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_darknet53_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_darknet53_270e_coco.yml) |
| MobileNet-V1 | 608 | 8 | 270e | ---- | 28.8 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_coco.yml) |
| MobileNet-V1 | 416 | 8 | 270e | ---- | 28.7 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_coco.yml) |
| MobileNet-V1 | 320 | 8 | 270e | ---- | 26.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_coco.yml) |
| MobileNet-V3 | 608 | 8 | 270e | ---- | 31.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_coco.yml) |
| MobileNet-V3 | 416 | 8 | 270e | ---- | 29.7 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_coco.yml) |
| MobileNet-V3 | 320 | 8 | 270e | ---- | 26.9 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_coco.yml) |
### YOLOv3 on Pasacl VOC
| 骨架网络 | 输入尺寸 | 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 |
| :----------- | :--: | :-----: | :-----: |:------------: |:----: | :-------: | :----: |
| MobileNet-V1 | 608 | 8 | 270e | - | 75.1 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_voc.yml) |
| MobileNet-V1 | 416 | 8 | 270e | - | 76.1 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_voc.yml) |
| MobileNet-V1 | 320 | 8 | 270e | - | 73.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_voc.yml) |
| MobileNet-V3 | 608 | 8 | 270e | - | 79.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_voc.yml) |
| MobileNet-V3 | 416 | 8 | 270e | - | 78.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_voc.yml) |
| MobileNet-V3 | 320 | 8 | 270e | - | 76.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_voc.yml) |
**注意:** YOLOv3均使用8GPU训练,训练270个epoch
## Citations
```
@misc{redmon2018yolov3,
title={YOLOv3: An Incremental Improvement},
author={Joseph Redmon and Ali Farhadi},
year={2018},
eprint={1804.02767},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
architecture: YOLOv3 architecture: YOLOv3
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/DarkNet53_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/DarkNet53_pretrained.tar
weights: output/yolov3_darknet/model_final
use_fine_grained_loss: false use_fine_grained_loss: false
load_static_weights: True load_static_weights: True
norm_type: sync_bn norm_type: sync_bn
......
architecture: YOLOv3 architecture: YOLOv3
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_pretrained.tar
weights: output/yolov3_mobilenet_v1/model_final
load_static_weights: True load_static_weights: True
norm_type: sync_bn norm_type: sync_bn
......
architecture: YOLOv3 architecture: YOLOv3
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_large_x1_0_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_large_x1_0_pretrained.tar
weights: output/yolov3_mobilenet_v3_large/model_final
load_static_weights: True load_static_weights: True
norm_type: sync_bn norm_type: sync_bn
......
architecture: YOLOv3 architecture: YOLOv3
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_small_x1_0_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_small_x1_0_pretrained.tar
weights: output/yolov3_mobilenet_v3_small/model_final
load_static_weights: True load_static_weights: True
norm_type: sync_bn norm_type: sync_bn
......
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_darknet53.yml',
'_base_/yolov3_reader.yml',
]
weights: output/yolov3_darknet53_coco/model_final
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_mobilenet_v1.yml',
'_base_/yolov3_reader.yml',
]
weights: output/yolov3_mobilenet_v1_coco/model_final
_BASE_: [ _BASE_: [
'./_base_/models/yolov3_mobilenet_v1.yml', '../datasets/voc.yml',
'./_base_/optimizers/yolov3_270e.yml', '../runtime.yml',
'./_base_/datasets/voc.yml', '_base_/optimizer_270e.yml',
'./_base_/readers/yolov3_reader.yml', '_base_/yolov3_mobilenet_v1.yml',
'./_base_/runtime.yml', '_base_/yolov3_reader.yml',
] ]
weights: output/yolov3_mobilenet_v1_voc/model_final
TrainReader: TrainReader:
inputs_def: inputs_def:
......
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_mobilenet_v3_large.yml',
'_base_/yolov3_reader.yml',
]
weights: output/yolov3_mobilenet_v3_large_coco/model_final
_BASE_: [ _BASE_: [
'./_base_/models/yolov3_mobilenet_v3_large.yml', '../datasets/voc.yml',
'./_base_/optimizers/yolov3_270e.yml', '../runtime.yml',
'./_base_/datasets/voc.yml', '_base_/optimizer_270e.yml',
'./_base_/readers/yolov3_reader.yml', '_base_/yolov3_mobilenet_v3_large.yml',
'./_base_/runtime.yml', '_base_/yolov3_reader.yml',
] ]
weights: output/yolov3_mobilenet_v3_large_voc/model_final
TrainReader: TrainReader:
inputs_def: inputs_def:
......
_BASE_: [
'./_base_/models/yolov3_darknet53.yml',
'./_base_/optimizers/yolov3_270e.yml',
'./_base_/datasets/coco_detection.yml',
'./_base_/readers/yolov3_reader.yml',
'./_base_/runtime.yml',
]
_BASE_: [
'./_base_/models/yolov3_mobilenet_v1.yml',
'./_base_/optimizers/yolov3_270e.yml',
'./_base_/datasets/coco_detection.yml',
'./_base_/readers/yolov3_reader.yml',
'./_base_/runtime.yml',
]
_BASE_: [
'./_base_/models/yolov3_mobilenet_v3_large.yml',
'./_base_/optimizers/yolov3_270e.yml',
'./_base_/datasets/coco_detection.yml',
'./_base_/readers/yolov3_reader.yml',
'./_base_/runtime.yml',
]
...@@ -23,67 +23,35 @@ ...@@ -23,67 +23,35 @@
## ImageNet预训练模型 ## ImageNet预训练模型
Paddle提供基于ImageNet的骨架网络预训练模型。所有预训练模型均通过标准的Imagenet-1k数据集训练得到[下载链接](https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/image_classification#supported-models-and-performances) Paddle提供基于ImageNet的骨架网络预训练模型。所有预训练模型均通过标准的Imagenet-1k数据集训练得到,ResNet和MobileNet等是采用余弦学习率调整策略或SSLD知识蒸馏训练得到的高精度预训练模型,可在[PaddleClas](https://github.com/PaddlePaddle/PaddleClas)查看模型细节。
- 注:ResNet50模型通过余弦学习率调整策略训练得到。[ResNet50下载链接](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet18_pretrained.tar),
[ResNet50_vd下载链接](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_pretrained.tar)
## 基线 ## 基线
### Faster & Mask R-CNN ### Faster R-CNN
| 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 | 请参考[Faster R-CNN](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/faster_rcnn/)
| :------------------- | :------------- | :-----: | :-----: | :------------: | :-----: | :-----: | :-----------------------------------------------------: | :-----: |
| ResNet50 | Faster | 1 | 1x | ---- | 35.1 | - | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/faster_rcnn_r50_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/faster_rcnn_r50_1x_coco.yml) |
| ResNet50-FPN | Faster | 1 | 1x | ---- | 37.0 | - | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/faster_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/faster_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50 | Mask | 1 | 1x | ---- | 36.4 | 31.9 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/mask_rcnn_r50_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/mask_rcnn_r50_1x_coco.yml) |
| ResNet50-FPN | Mask | 1 | 1x | ---- | 38.3 | 34.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/mask_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/mask_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50-FPN | Cascade Faster | 1 | 1x | ---- | 41.1 | - | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/cascade_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/cascade_faster_rcnn_r50_fpn_1x_coco.yml) |
| ResNet50-FPN | Cascade Mask | 1 | 1x | ---- | 41.6 | 35.3 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/cascade_mask_rcnn_r50_fpn_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/cascade_mask_rcnn_r50_fpn_1x_coco.yml) |
### Mask R-CNN
### YOLOv3 on COCO 请参考[Mask R-CNN](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/mask_rcnn/)
| 骨架网络 | 输入尺寸 | 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 | ### Cascade R-CNN
| :------------------- | :------- | :-----: | :-----: | :------------: | :-----: | :-----------------------------------------------------: | :-----: |
| DarkNet53(paper) | 608 | 8 | 270e | ---- | 33.0 | - | - |
| DarkNet53(paper) | 416 | 8 | 270e | ---- | 31.0 | - | - |
| DarkNet53(paper) | 320 | 8 | 270e | ---- | 28.2 | - | - |
| DarkNet53 | 608 | 8 | 270e | ---- | 39.0 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_darknet53_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_darknet53_270e_coco.yml) |
| DarkNet53 | 416 | 8 | 270e | ---- | 37.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_darknet53_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_darknet53_270e_coco.yml) |
| DarkNet53 | 320 | 8 | 270e | ---- | 34.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_darknet53_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_darknet53_270e_coco.yml) |
| MobileNet-V1 | 608 | 8 | 270e | ---- | 28.8 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_coco.yml) |
| MobileNet-V1 | 416 | 8 | 270e | ---- | 28.7 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_coco.yml) |
| MobileNet-V1 | 320 | 8 | 270e | ---- | 26.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_coco.yml) |
| MobileNet-V3 | 608 | 8 | 270e | ---- | 31.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_coco.yml) |
| MobileNet-V3 | 416 | 8 | 270e | ---- | 29.7 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_coco.yml) |
| MobileNet-V3 | 320 | 8 | 270e | ---- | 26.9 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_coco.yml) |
### YOLOv3 on Pasacl VOC 请参考[Cascade R-CNN](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/cascade_rcnn/)
| 骨架网络 | 输入尺寸 | 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 | ### YOLOv3
| :----------- | :--: | :-----: | :-----: |:------------: |:----: | :-------: | :----: |
| MobileNet-V1 | 608 | 8 | 270e | - | 75.1 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_voc.yml) |
| MobileNet-V1 | 416 | 8 | 270e | - | 76.1 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_voc.yml) |
| MobileNet-V1 | 320 | 8 | 270e | - | 73.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v1_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v1_270e_voc.yml) |
| MobileNet-V3 | 608 | 8 | 270e | - | 79.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_voc.yml) |
| MobileNet-V3 | 416 | 8 | 270e | - | 78.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_voc.yml) |
| MobileNet-V3 | 320 | 8 | 270e | - | 76.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/yolov3_mobilenet_v3_large_270e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3_mobilenet_v3_large_270e_voc.yml) |
**注意:** YOLOv3均使用8GPU训练,训练270个epoch 请参考[YOLOv3](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/yolov3/)
### SSD on Pascal VOC ### SSD
| 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 | 请参考[SSD](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/ssd/)
| :-------------- | :------------- | :-----: | :-----: | :------------: | :-----: | :-----------------------------------------------------: | :-----: |
| VGG | SSD | 8 | 240e | ---- | 78.2 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/dygraph/ssd_vgg16_300_240e_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/ssd_vgg16_300_240e_voc.yml) |
**注意:** SSD使用4GPU训练,训练240个epoch
### FCOS ### FCOS
请参考[fcos](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/fcos/) 请参考[FCOS](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/fcos/)
### SOLOv2 ### SOLOv2
请参考[solov2](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/solov2/) 请参考[SOLOv2](https://github.com/PaddlePaddle/PaddleDetection/tree/dygraph/configs/solov2/)
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