Skip to content

P
PaddleDetection

s920243400 / PaddleDetection
与 Fork 源项目一致

Fork自 PaddlePaddle / PaddleDetection

通知 2
Star 0
Fork 0
P
PaddleDetection
未验证 提交 c0647323 编写于 作者: F Feng Ni 提交者： GitHub
浏览文件
操作

[semi_det] fix save_model and rename folder (#7433) 

* fix save_model epoch_id and rename folder

* fix use_simple_ema and docs

* fix docs, test=document_fix
上级 3529e38f
显示空白变更内容
内联 并排
Showing with 300 addition and 30 deletion
configs/ssod/README.md → configs/semi_det/README.md
浏览文件 @ c0647323
简体中文 | [English](README_en.md) 简体中文 | [English](README_en.md)
# Semi-Supervised Object Detection (SSOD) 半监督目标检测 # Semi-Supervised Detection (Semi DET) 半监督检测
## 内容 ## 内容
- [简介](#简介) - [简介](#简介)
...@@ -34,12 +34,12 @@ ...@@ -34,12 +34,12 @@
### [DenseTeacher](denseteacher) ### [DenseTeacher](denseteacher)
| 模型 | 基础检测器 | 监督数据比例 | Sup mAP<sup>val<br>0.5:0.95 | Semi mAP<sup>val<br>0.5:0.95 | Semi Epochs (Iters) | 模型下载 | 配置文件 | | 模型 | 监督数据比例 | Sup Baseline | Sup Epochs (Iters) | Sup mAP<sup>val<br>0.5:0.95 | Semi mAP<sup>val<br>0.5:0.95 | Semi Epochs (Iters) | 模型下载 | 配置文件 |
| :------------: | :---------------------: | :-----------: | :-------------------------: |:---------------------------: |:--------------------: | :-------: |:---------: | | :------------: | :---------: | :---------------------: | :---------------------: |:---------------------------: |:----------------------------: | :------------------: |:--------: |:----------: |
| DenseTeacher | [FCOS ResNet50-FPN](./baseline/fcos_r50_fpn_2x_coco_sup005.yml) | 5% | 21.3 | 30.6 | 240 (87120) | [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi005.pdparams) | [config](denseteacher/denseteacher_fcos_r50_fpn_coco_semi005.yml) | | DenseTeacher-FCOS | 5% | [sup_config](./baseline/fcos_r50_fpn_2x_coco_sup005.yml) | 24 (8712) | 21.3 | **30.6** | 240 (87120) | [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi005.pdparams) | [config](denseteacher/denseteacher_fcos_r50_fpn_coco_semi005.yml) |
| DenseTeacher | [FCOS ResNet50-FPN](./baseline/fcos_r50_fpn_2x_coco_sup010.yml) | 10%| 26.3 | 35.1 | 240 (174240)| [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010.pdparams) | [config](denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml) | | DenseTeacher-FCOS | 5% | [sup_config](./baseline/fcos_r50_fpn_2x_coco_sup010.yml) | 24 (17424) | 26.3 | **35.1** | 240 (174240) | [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010.pdparams) | [config](denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml) |
| DenseTeacher(LSJ)| [FCOS ResNet50-FPN](./baseline/fcos_r50_fpn_2x_coco_sup010.yml) |10%| 26.3 | 37.1 | 240 (174240)| [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010_lsj.pdparams) | [config](denseteacher/denseteacher_fcos_r50_fpn_coco_semi010_lsj.yml) | | DenseTeacher-FCOS(LSJ)| 5% | [sup_config](./baseline/fcos_r50_fpn_2x_coco_sup010.yml) | 24 (17424) | 26.3 | **37.1(LSJ)** | 240 (174240) | [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010_lsj.pdparams) | [config](denseteacher/denseteacher_fcos_r50_fpn_coco_semi010_lsj.yml) |
| DenseTeacher | [FCOS ResNet50-FPN](./../fcos/fcos_r50_fpn_iou_multiscale_2x_coco.ymll) |full| 42.6 | - | 36 (263844)| [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_full.pdparams) | [config](denseteacher/denseteacher_fcos_r50_fpn_coco_full.yml) | | DenseTeacher-FCOS |100%(full)| [sup_config](./../fcos/fcos_r50_fpn_iou_multiscale_2x_coco.ymll) | 24 (175896) | 42.6 | **44.2** | 24 (175896)| [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_full.pdparams) | [config](denseteacher/denseteacher_fcos_r50_fpn_coco_full.yml) |
## 半监督数据集准备 ## 半监督数据集准备
...@@ -215,7 +215,7 @@ use_warmup: &use_warmup True ...@@ -215,7 +215,7 @@ use_warmup: &use_warmup True
### 全局配置 ### 全局配置
需要在配置文件中添加如下全局配置： 需要在配置文件中添加如下全局配置，并且注意 DenseTeacher 模型需要使用`use_simple_ema: True`而不是`use_ema: True`：
```python ```python
### global config ### global config
...@@ -249,9 +249,9 @@ _BASE_: [ ...@@ -249,9 +249,9 @@ _BASE_: [
### 数据增强配置 ### 数据增强配置
构建半监督训练集的Reader，需要在原先`TrainReader`的基础上，新增加`weak_aug`,`strong_aug`,`sup_batch_transforms`和`unsup_batch_transforms`，并且需要注意： 构建半监督训练集的Reader，需要在原先`TrainReader`的基础上，新增加`weak_aug`,`strong_aug`,`sup_batch_transforms`和`unsup_batch_transforms`，并且需要注意：
- **如果有`NormalizeImage`，需要单独从`sample_transforms`中抽出来放在`weak_aug`和`strong_aug`中； - 如果有`NormalizeImage`，需要单独从`sample_transforms`中抽出来放在`weak_aug`和`strong_aug`中；
- `sample_transforms`为**公用的基础数据增强**； - `sample_transforms`为**公用的基础数据增强**；
- 完整的弱数据增强为``sample_transforms + weak_aug`，完整的强数据增强为`sample_transforms + strong_aug`； - 完整的弱数据增强为`sample_transforms + weak_aug`，完整的强数据增强为`sample_transforms + strong_aug`；
如以下所示: 如以下所示:
...@@ -302,8 +302,6 @@ SemiTrainReader: ...@@ -302,8 +302,6 @@ SemiTrainReader:
object_sizes_boundary: [64, 128, 256, 512] object_sizes_boundary: [64, 128, 256, 512]
center_sampling_radius: 1.5 center_sampling_radius: 1.5
downsample_ratios: [8, 16, 32, 64, 128] downsample_ratios: [8, 16, 32, 64, 128]
num_shift: 0.5
multiply_strides_reg_targets: False
norm_reg_targets: True norm_reg_targets: True
unsup_batch_transforms: unsup_batch_transforms:
- Permute: {} - Permute: {}
...@@ -351,22 +349,22 @@ OptimizerBuilder: ...@@ -351,22 +349,22 @@ OptimizerBuilder:
```bash ```bash
# 单卡训练 (不推荐，需按线性比例相应地调整学习率) # 单卡训练 (不推荐，需按线性比例相应地调整学习率)
CUDA_VISIBLE_DEVICES=0 python tools/train.py -c ssod/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml --eval CUDA_VISIBLE_DEVICES=0 python tools/train.py -c configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml --eval
# 多卡训练 # 多卡训练
python -m paddle.distributed.launch --log_dir=denseteacher_fcos_semi010/ --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ssod/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml --eval python -m paddle.distributed.launch --log_dir=denseteacher_fcos_semi010/ --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml --eval
``` ```
### 评估 ### 评估
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c ssod/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=output/denseteacher_fcos_r50_fpn_coco_semi010/model_final.pdparams CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=output/denseteacher_fcos_r50_fpn_coco_semi010/model_final.pdparams
``` ```
### 预测 ### 预测
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ssod/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=output/denseteacher_fcos_r50_fpn_coco_semi010/model_final.pdparams --infer_img=demo/000000014439.jpg CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=output/denseteacher_fcos_r50_fpn_coco_semi010/model_final.pdparams --infer_img=demo/000000014439.jpg
``` ```
### 部署 ### 部署
...@@ -375,7 +373,7 @@ CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ssod/denseteacher/denseteacher_f ...@@ -375,7 +373,7 @@ CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ssod/denseteacher/denseteacher_f
```bash ```bash
# 导出模型 # 导出模型
CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ssod/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010.pdparams CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010.pdparams
# 导出权重预测 # 导出权重预测
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/denseteacher_fcos_r50_fpn_coco_semi010 --image_file=demo/000000014439_640x640.jpg --device=GPU CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/denseteacher_fcos_r50_fpn_coco_semi010 --image_file=demo/000000014439_640x640.jpg --device=GPU
......
configs/ssod/baseline/README.md → configs/semi_det/baseline/README.md
浏览文件 @ c0647323
...@@ -62,7 +62,7 @@ ...@@ -62,7 +62,7 @@
将以下命令写在一个脚本文件里如```run.sh```，一键运行命令为：```sh run.sh```，也可命令行一句句去运行： 将以下命令写在一个脚本文件里如```run.sh```，一键运行命令为：```sh run.sh```，也可命令行一句句去运行：
```bash ```bash
model_type=ssod/baseline model_type=semi_det/baseline
job_name=ppyoloe_plus_crn_s_80e_coco_sup010 # 可修改，如 fcos_r50_fpn_2x_coco_sup010 job_name=ppyoloe_plus_crn_s_80e_coco_sup010 # 可修改，如 fcos_r50_fpn_2x_coco_sup010
config=configs/${model_type}/${job_name}.yml config=configs/${model_type}/${job_name}.yml
......
configs/ssod/denseteacher/README.md → configs/semi_det/denseteacher/README.md
浏览文件 @ c0647323
简体中文 | [English](README_en.md) 简体中文 | [English](README_en.md)
# DenseTeacher (Dense Teacher: Dense Pseudo-Labels for Semi-supervised Object Detection) # Dense Teacher: Dense Pseudo-Labels for Semi-supervised Object Detection
## 模型库 ## 模型库
| 模型 | 基础检测器 | 监督数据比例 | Sup mAP<sup>val<br>0.5:0.95 | Semi mAP<sup>val<br>0.5:0.95 | Semi Epochs (Iters) | 模型下载 | 配置文件 | | 模型 | 监督数据比例 | Sup Baseline | Sup Epochs (Iters) | Sup mAP<sup>val<br>0.5:0.95 | Semi mAP<sup>val<br>0.5:0.95 | Semi Epochs (Iters) | 模型下载 | 配置文件 |
| :------------: | :---------------------: | :-----------: | :-------------------------: |:---------------------------: |:--------------------: | :-------: |:---------: | | :------------: | :---------: | :---------------------: | :---------------------: |:---------------------------: |:----------------------------: | :------------------: |:--------: |:----------: |
| DenseTeacher | [FCOS R50-FPN](../baseline/fcos_r50_fpn_2x_coco_sup005.yml) | 5% | 21.3 | 30.6 | 240 (87120) | [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi005.pdparams) | [config](./denseteacher_fcos_r50_fpn_coco_semi005.yml) | | DenseTeacher-FCOS | 5% | [sup_config](../baseline/fcos_r50_fpn_2x_coco_sup005.yml) | 24 (8712) | 21.3 | **30.6** | 240 (87120) | [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi005.pdparams) | [config](./denseteacher_fcos_r50_fpn_coco_semi005.yml) |
| DenseTeacher | [FCOS R50-FPN](../baseline/fcos_r50_fpn_2x_coco_sup010.yml) | 10%| 26.3 | 35.1 | 240 (174240)| [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010.pdparams) | [config](./denseteacher_fcos_r50_fpn_coco_semi010.yml) | | DenseTeacher-FCOS | 5% | [sup_config](../baseline/fcos_r50_fpn_2x_coco_sup010.yml) | 24 (17424) | 26.3 | **35.1** | 240 (174240) | [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010.pdparams) | [config](./denseteacher_fcos_r50_fpn_coco_semi010.yml) |
| DenseTeacher(LSJ)| [FCOS R50-FPN](../baseline/fcos_r50_fpn_2x_coco_sup010.yml)| 10%| 26.3 | 37.1 | 240 (174240)| [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010_lsj.pdparams) | [config](./denseteacher_fcos_r50_fpn_coco_semi010_lsj.yml) | | DenseTeacher-FCOS(LSJ)| 5% | [sup_config](../baseline/fcos_r50_fpn_2x_coco_sup010.yml) | 24 (17424) | 26.3 | **37.1(LSJ)** | 240 (174240) | [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010_lsj.pdparams) | [config](./denseteacher_fcos_r50_fpn_coco_semi010_lsj.yml) |
| DenseTeacher | [FCOS R50-FPN](../../fcos/fcos_r50_fpn_iou_multiscale_2x_coco.ymll) |full| 42.6 | - | 36 (263844)| [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_full.pdparams) | [config](./denseteacher_fcos_r50_fpn_coco_full.yml) | | DenseTeacher-FCOS |100%(full)| [sup_config](../../fcos/fcos_r50_fpn_iou_multiscale_2x_coco.ymll) | 24 (175896) | 42.6 | **44.2** | 24 (175896)| [download](https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_full.pdparams) | [config](./denseteacher_fcos_r50_fpn_coco_full.yml) |
**注意:** **注意:**
- 以上模型训练默认使用8 GPUs，监督数据总batch_size默认为16，无监督数据总batch_size默认也为16，默认初始学习率为0.01。如果改动了总batch_size，请按线性比例相应地调整学习率； - 以上模型训练默认使用8 GPUs，监督数据总batch_size默认为16，无监督数据总batch_size默认也为16，默认初始学习率为0.01。如果改动了总batch_size，请按线性比例相应地调整学习率；
- **监督数据比例**是指使用的有标签COCO数据集占 COCO train2017 全量训练集的百分比，使用的无标签COCO数据集一般也是相同比例，但具体图片和有标签数据的图片不重合；
- `Semi Epochs (Iters)`表示**半监督训练**的模型的 Epochs (Iters)，如果使用**自定义数据集**，需自行根据Iters换算到对应的Epochs调整，最好保证总Iters 和COCO数据集的设置较为接近； - `Semi Epochs (Iters)`表示**半监督训练**的模型的 Epochs (Iters)，如果使用**自定义数据集**，需自行根据Iters换算到对应的Epochs调整，最好保证总Iters 和COCO数据集的设置较为接近；
- `Sup mAP`是**只使用有监督数据训练**的模型的精度，请参照**基础检测器的配置文件** 和 [baseline](../baseline)； - `Sup mAP`是**只使用有监督数据训练**的模型的精度，请参照**基础检测器的配置文件** 和 [baseline](../baseline)；
- `Semi mAP`是**半监督训练**的模型的精度，模型下载和配置文件的链接均为**半监督模型**； - `Semi mAP`是**半监督训练**的模型的精度，模型下载和配置文件的链接均为**半监督模型**；
- `LSJ`表示 large-scale jittering，表示更大范围的多尺度训练，可进一步提升精度，但训练速度也会变慢； - `LSJ`表示 **large-scale jittering**，表示使用更大范围的多尺度训练，可进一步提升精度，但训练速度也会变慢；
- 半监督检测的配置和使用，请参照[文档](../README.md/#半监督检测配置)； - 半监督检测的配置和使用，请参照[文档](../README.md/#半监督检测配置)；
- `Dense Teacher`原文使用`R50-va-caffe`预训练，PaddleDetection中默认使用`R50-vb`预训练，如果使用`R50-vd`结合[SSLD](../../../docs/feature_models/SSLD_PRETRAINED_MODEL.md)的预训练模型，可进一步显著提升检测精度，如：
```
pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams
ResNet:
depth: 50
variant: d
norm_type: bn
freeze_at: 0
return_idx: [0,1,2,3]
num_stages: 4
lr_mult_list: [0.05, 0.05, 0.1, 0.15]
```
## 使用说明 ## 使用说明
...@@ -29,22 +42,22 @@ ...@@ -29,22 +42,22 @@
```bash ```bash
# 单卡训练 (不推荐，需按线性比例相应地调整学习率) # 单卡训练 (不推荐，需按线性比例相应地调整学习率)
CUDA_VISIBLE_DEVICES=0 python tools/train.py -c ssod/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml --eval CUDA_VISIBLE_DEVICES=0 python tools/train.py -c configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml --eval
# 多卡训练 # 多卡训练
python -m paddle.distributed.launch --log_dir=denseteacher_fcos_semi010/ --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ssod/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml --eval python -m paddle.distributed.launch --log_dir=denseteacher_fcos_semi010/ --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml --eval
``` ```
### 评估 ### 评估
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c ssod/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=output/denseteacher_fcos_r50_fpn_coco_semi010/model_final.pdparams CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=output/denseteacher_fcos_r50_fpn_coco_semi010/model_final.pdparams
``` ```
### 预测 ### 预测
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ssod/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=output/denseteacher_fcos_r50_fpn_coco_semi010/model_final.pdparams --infer_img=demo/000000014439.jpg CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=output/denseteacher_fcos_r50_fpn_coco_semi010/model_final.pdparams --infer_img=demo/000000014439.jpg
``` ```
### 部署 ### 部署
...@@ -53,7 +66,7 @@ CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ssod/denseteacher/denseteacher_f ...@@ -53,7 +66,7 @@ CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ssod/denseteacher/denseteacher_f
```bash ```bash
# 导出模型 # 导出模型
CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ssod/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010.pdparams CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_semi010.yml -o weights=https://paddledet.bj.bcebos.com/models/denseteacher_fcos_r50_fpn_coco_semi010.pdparams
# 导出权重预测 # 导出权重预测
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/denseteacher_fcos_r50_fpn_coco_semi010 --image_file=demo/000000014439_640x640.jpg --device=GPU CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/denseteacher_fcos_r50_fpn_coco_semi010 --image_file=demo/000000014439_640x640.jpg --device=GPU
......
configs/semi_det/denseteacher/denseteacher_fcos_r50_fpn_coco_full.yml 0 → 100644
浏览文件 @ c0647323
_BASE_: [
'denseteacher_fcos_r50_fpn_coco_semi010.yml',
'../_base_/coco_detection_full.yml',
]
log_iter: 100
snapshot_epoch: 2
epochs: &epochs 24
weights: output/denseteacher_fcos_r50_fpn_coco_full/model_final
### pretrain and warmup config, choose one and coment another
pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_cos_pretrained.pdparams
semi_start_iters: 5000
ema_start_iters: 3000
use_warmup: &use_warmup True
### global config
use_simple_ema: True
ema_decay: 0.9996
ssod_method: DenseTeacher
DenseTeacher:
train_cfg:
sup_weight: 1.0
unsup_weight: 1.0
loss_weight: {distill_loss_cls: 2.0, distill_loss_box: 1.0, distill_loss_quality: 1.0}
concat_sup_data: True
suppress: linear
ratio: 0.01
gamma: 2.0
test_cfg:
inference_on: teacher
### reader config
worker_num: 2
SemiTrainReader:
sample_transforms:
- Decode: {}
- RandomResize: {target_size: [[640, 1333], [672, 1333], [704, 1333], [736, 1333], [768, 1333], [800, 1333]], keep_ratio: True, interp: 1}
- RandomFlip: {}
weak_aug:
- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: true}
strong_aug:
- StrongAugImage: {transforms: [
RandomColorJitter: {prob: 0.8, brightness: 0.4, contrast: 0.4, saturation: 0.4, hue: 0.1},
RandomErasingCrop: {},
RandomGaussianBlur: {prob: 0.5, sigma: [0.1, 2.0]},
RandomGrayscale: {prob: 0.2},
]}
- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: true}
sup_batch_transforms:
- Permute: {}
- PadBatch: {pad_to_stride: 32}
- Gt2FCOSTarget:
object_sizes_boundary: [64, 128, 256, 512]
center_sampling_radius: 1.5
downsample_ratios: [8, 16, 32, 64, 128]
norm_reg_targets: True
unsup_batch_transforms:
- Permute: {}
- PadBatch: {pad_to_stride: 32}
sup_batch_size: 2
unsup_batch_size: 2
shuffle: True
drop_last: True
EvalReader:
sample_transforms:
- Decode: {}
- Resize: {target_size: [800, 1333], keep_ratio: True, interp: 1}
- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
- Permute: {}
batch_transforms:
- PadBatch: {pad_to_stride: 32}
batch_size: 1
TestReader:
sample_transforms:
- Decode: {}
- Resize: {target_size: [800, 1333], keep_ratio: True, interp: 1}
- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
- Permute: {}
batch_transforms:
- PadBatch: {pad_to_stride: 32}
batch_size: 1
fuse_normalize: True
### model config
architecture: FCOS
FCOS:
backbone: ResNet
neck: FPN
fcos_head: FCOSHead
ResNet:
depth: 50
variant: 'b'
norm_type: bn
freeze_at: 0 # res2
return_idx: [1, 2, 3]
num_stages: 4
FPN:
out_channel: 256
spatial_scales: [0.125, 0.0625, 0.03125]
extra_stage: 2
has_extra_convs: True
use_c5: False
FCOSHead:
fcos_feat:
name: FCOSFeat
feat_in: 256
feat_out: 256
num_convs: 4
norm_type: "gn"
use_dcn: False
fpn_stride: [8, 16, 32, 64, 128]
prior_prob: 0.01
norm_reg_targets: True
centerness_on_reg: True
fcos_loss:
name: FCOSLoss
loss_alpha: 0.25
loss_gamma: 2.0
iou_loss_type: "giou"
reg_weights: 1.0
quality: "iou"
nms:
name: MultiClassNMS
nms_top_k: 1000
keep_top_k: 100
score_threshold: 0.05
nms_threshold: 0.6
### other config
epoch: *epochs
LearningRate:
base_lr: 0.01
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones: [*epochs]
use_warmup: *use_warmup
- !LinearWarmup
start_factor: 0.001
steps: 1000
OptimizerBuilder:
optimizer:
momentum: 0.9
type: Momentum
regularizer:
factor: 0.0001
type: L2
clip_grad_by_value: 1.0
ppdet/engine/callbacks.py
浏览文件 @ c0647323
...@@ -159,7 +159,6 @@ class LogPrinter(Callback): ...@@ -159,7 +159,6 @@ class LogPrinter(Callback):
class Checkpointer(Callback): class Checkpointer(Callback):
def __init__(self, model): def __init__(self, model):
super(Checkpointer, self).__init__(model) super(Checkpointer, self).__init__(model)
cfg = self.model.cfg
self.best_ap = -1000. self.best_ap = -1000.
self.save_dir = os.path.join(self.model.cfg.save_dir, self.save_dir = os.path.join(self.model.cfg.save_dir,
self.model.cfg.filename) self.model.cfg.filename)
...@@ -210,6 +209,9 @@ class Checkpointer(Callback): ...@@ -210,6 +209,9 @@ class Checkpointer(Callback):
key, eval_func, abs(self.best_ap))) key, eval_func, abs(self.best_ap)))
if weight: if weight:
if self.model.use_ema: if self.model.use_ema:
exchange_save_model = status.get('exchange_save_model',
False)
if not exchange_save_model:
# save model and ema_model # save model and ema_model
save_model( save_model(
status['weight'], status['weight'],
...@@ -218,6 +220,21 @@ class Checkpointer(Callback): ...@@ -218,6 +220,21 @@ class Checkpointer(Callback):
save_name, save_name,
epoch_id + 1, epoch_id + 1,
ema_model=weight) ema_model=weight)
else:
# save model(student model) and ema_model(teacher model)
# in DenseTeacher SSOD, the teacher model will be higher,
# so exchange when saving pdparams
student_model = status['weight'] # model
teacher_model = weight # ema_model
save_model(
teacher_model,
self.model.optimizer,
self.save_dir,
save_name,
epoch_id + 1,
ema_model=student_model)
del teacher_model
del student_model
else: else:
save_model(weight, self.model.optimizer, self.save_dir, save_model(weight, self.model.optimizer, self.save_dir,
save_name, epoch_id + 1) save_name, epoch_id + 1)
......
ppdet/engine/trainer_ssod.py
浏览文件 @ c0647323
...@@ -191,8 +191,11 @@ class Trainer_DenseTeacher(Trainer): ...@@ -191,8 +191,11 @@ class Trainer_DenseTeacher(Trainer):
self.status.update({ self.status.update({
'epoch_id': self.start_epoch, 'epoch_id': self.start_epoch,
'step_id': 0, 'step_id': 0,
'steps_per_epoch': len(self.loader) 'steps_per_epoch': len(self.loader),
'exchange_save_model': True,
}) })
# Note: exchange_save_model
# in DenseTeacher SSOD, the teacher model will be higher, so exchange when saving pdparams
self.status['batch_time'] = stats.SmoothedValue( self.status['batch_time'] = stats.SmoothedValue(
self.cfg.log_iter, fmt='{avg:.4f}') self.cfg.log_iter, fmt='{avg:.4f}')
...@@ -254,8 +257,12 @@ class Trainer_DenseTeacher(Trainer): ...@@ -254,8 +257,12 @@ class Trainer_DenseTeacher(Trainer):
data_sup_w, data_sup_s = align_weak_strong_shape(data_sup_w, data_sup_w, data_sup_s = align_weak_strong_shape(data_sup_w,
data_sup_s) data_sup_s)
data_sup_w['epoch_id'] = epoch_id
data_sup_s['epoch_id'] = epoch_id
if concat_sup_data: if concat_sup_data:
for k, v in data_sup_s.items(): for k, v in data_sup_s.items():
if k in ['epoch_id']:
continue
data_sup_s[k] = paddle.concat([v, data_sup_w[k]]) data_sup_s[k] = paddle.concat([v, data_sup_w[k]])
loss_dict_sup = self.model(data_sup_s) loss_dict_sup = self.model(data_sup_s)
else: else:
...@@ -264,8 +271,6 @@ class Trainer_DenseTeacher(Trainer): ...@@ -264,8 +271,6 @@ class Trainer_DenseTeacher(Trainer):
for k, v in loss_dict_sup_w.items(): for k, v in loss_dict_sup_w.items():
loss_dict_sup[k] = (loss_dict_sup[k] + v) * 0.5 loss_dict_sup[k] = (loss_dict_sup[k] + v) * 0.5
data_sup_w['epoch_id'] = epoch_id
data_sup_s['epoch_id'] = epoch_id
losses_sup = loss_dict_sup['loss'] * train_cfg['sup_weight'] losses_sup = loss_dict_sup['loss'] * train_cfg['sup_weight']
losses_sup.backward() losses_sup.backward()
......
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