Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
wh 644e35cd25 | 2 years ago | |
---|---|---|
scripts | 2 years ago | |
src | 2 years ago | |
CHANGELOG.md | 2 years ago | |
LICENSE | 2 years ago | |
README.md | 2 years ago | |
creat_Midrecord.py | 2 years ago | |
eval.py | 2 years ago | |
export.py | 2 years ago | |
requirements.txt | 2 years ago | |
train.py | 2 years ago |
YOLACT提出的实时实例分割算法在2020年被作者扩展为YOLACT++:更好的实时实例分割。在COCO的test dev数据集上达到34.1mAP。YOLACT++在保证实时性(大于或等于30fps)的前提下,对原版的YOLACT做出几点改进,大幅提升了mAP。
在YOLACT++ 中,将ResNet的C3-C5中的各个标准3x3卷积换成3x3可变性卷积,但没有使用堆叠的可变形卷积模块,因为延迟太高。优化了Prediction Head分支,由于YOLACT是anchor-based的,所以对anchor设计进行优化。YOLACT++受MS R-CNN的启发,高质量的mask并不一定就对应着高的分类置信度,所以在模型后添加了Mask Re-Scoring分支,该分支使用YOLACT生成的裁剪后的原型mask(未作阈值化)作为输入,输出对应每个类别的GT-mask的IoU。
论文: Bolya D, Zhou C, Xiao F, et al. YOLACT++: Better Real-time Instance Segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020
使用的数据集:COCO2017
├── Yolact++
├── train.py // 训练脚本
├── eval.py // 评估脚本
├── export.py // 导出模型的脚本
├── README.md
├── requirements.txt
├── scripts
├── run_standalone_train.sh // 单卡脚本
├── run_distribute_train.sh // 多卡脚本
├── src
├── __init__.py // 初始化文件
├── config.py // 参数配置文件
├── loss_monitor.py // 监视对loss训练是否正常
├── lr_schedule.py // 学习率
├── dataset.py // 创建数据集
├── network_define.py // yolact 训练网络封装器
├── yolact
├── __init__.py
├── yolactpp.py // 模型架构
├── layers
├── __init__.py
├── backbone_dcnV2.py // Backbone
├── fpn.py // FPN
├── protonet.py // Protonet
├── functions
├── __init__.py
├── detection_614.py
├── modules
├── __init__.py
├── loss_614.py // 损失函数计算
├── match_614.py // 训练时计算Iou与标签匹配
├── utils
├── __init__.py
├── functions.py // 生成网络
├── interpolate.py
├── ms_box_utils.py // 验证时计算Box Iou与编码和解码
run_standalone_train.sh
开始Yolact++模型的非分布式训练。bash run_standalone_train.sh DEVICE_ID
run_distribute_train.sh
开始Yolact++模型的分布式训练。bash run_distribute_train.sh RANK_TABLE_FILE DEVICE_NUMS
Parameters | |
---|---|
Model Version | Yolact++ |
Resource | CentOs 8.2; Ascend 910; CPU 2.60GHz, 192cores; Memory 755G |
MindSpore Version | 1.3.0 |
Dataset | COCO2017 |
Training Parameters | epoch = 300, batch_size = 8 |
Optimizer | Momentum |
Loss Function | semantic_segmentation_loss, ohem_conf_loss, mask_iou_loss, lincomb_mask_loss |
outputs | super-resolution pictures |
Accuracy | 0 |
Speed | 1pc(Ascend): 10000 ms/step |
Total time | 1pc: 270days |
Scripts | Yolact++ script |
Please check the official homepage.
Yolact plus 实例分割网络的mindspore实现。
Python Shell other
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》