论文：YOLACT: Real-time Instance Segmentation
使用 Paddle 复现
本项目基于paddlepaddle框架复现YOLACT，YOLACT是一种新的实例分割网络，由于没有使用两阶段方法中的pooling操作使得可以获得无损失的特征信息，并且在大目标的分割场景下性能表现更优秀
论文地址：
https://arxiv.org/pdf/1904.02689.pdf
参考项目：
https://github.com/dbolya/yolact
https://github.com/PaddlePaddle/PaddleDetection
https://github.com/PaddlePaddle/PaddleSeg
COCO test-dev2017
| Image Size | Backbone | mAP | download |
|---|---|---|---|
| 550 | Resnet50-FPN | 28.9 | 百度网盘 提取码: 6jjs |
COCO2017-完整数据集:
https://aistudio.baidu.com/aistudio/datasetdetail/97273
pip install -r requirments.txt
-
硬件：GPU、CPU
-
框架：
aistudio 默认 2.1 版本
Name: paddlepaddle-gpu
Version: 2.1.2.post101
修改训练数据配置文件
修改配置文件 data/config.py
dataset_base = Config({
'name': 'Base Dataset',
# Training images and annotations
'train_images': '/home/aistudio/train2017/', # 修改为训练数据文件夹
'train_info': 'path_to_annotation_file',
# Validation images and annotations.
'valid_images': '/home/aistudio/val2017/', # 修改为验证数据文件夹
'valid_info': 'path_to_annotation_file',
# Whether or not to load GT. If this is False, eval.py quantitative evaluation won't work.
'has_gt': True,
# A list of names for each of you classes.
'class_names': COCO_CLASSES,
# COCO class ids aren't sequential, so this is a bandage fix. If your ids aren't sequential,
# provide a map from category_id -> index in class_names + 1 (the +1 is there because it's 1-indexed).
# If not specified, this just assumes category ids start at 1 and increase sequentially.
'label_map': None
})
coco2017_dataset = dataset_base.copy({
'name': 'COCO 2017',
'train_info': '/home/aistudio/annotations/instances_train2017.json', # 修改为 train 据标注
'valid_info': '/home/aistudio/annotations/instances_val2017.json', # 修改为 val 数据标注
'label_map': COCO_LABEL_MAP
})
coco2017_testdev_dataset = dataset_base.copy({
'name': 'COCO 2017 Test-Dev',
'valid_images': '/home/aistudio/test2017/', # coco test-dev2017 测试数据集文件夹
'valid_info': '/home/aistudio/annotations/image_info_test-dev2017.json', # image_info_test-dev2017.json 文件
'has_gt': False,
'label_map': COCO_LABEL_MAP
})
预测
python eval.py --trained_model yolact_resnet50_54_800000.dpparams --output_coco_json --dataset=coco2017_testdev_dataset --cuda=True
result文件夹中mask_detections.json生成结果
训练
python train.py --config=yolact_resnet50_config --batch_size=8
├─data # 数据加载和配置
|--config.py # 配置文件
├─layers # 中间处理过程和 loss
|--modules
|--multibox_loss.py # 训练的 loss
├─logs # 训练日志 每 10 个 epoch 打印
├─utils # 工具包（计时、日志记录、数据增强）
|--weights # 模型保存目录
│--backbone.py # backbone(resnet 实现)
│--eval.py # 预测和评估
│--yolact.py # model 实现
│--README.md # 中文readme
│--requirement.txt # 依赖
│--train.py # 训练
下载 yolact_resnet50_54_800000.dpparams 并保存到 eval.py 相同目录下
python eval.py --trained_model yolact_resnet50_54_800000.dpparams --output_coco_json --dataset=coco2017_testdev_dataset --cuda=True
cd results
cp mask_detections.json detections_test-dev2017_yolact_results.json
zip detections_test-dev2017_yolact_results.zip detections_test-dev2017_yolact_results.json
在 coco 官网提交结果
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.289
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.477
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.301
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.093
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.301
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.460
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.273
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.395
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.407
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.188
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.439
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.594