OpenMMLabCamp icon indicating copy to clipboard operation
OpenMMLabCamp copied to clipboard
verified publisher icon open-mmlab

【AI实战营第二期】第四次作业提交07班

Open chg0901 opened this issue 2 years ago • 6 comments

作业：MMSeg 语义分割

背景：西瓜瓤、西瓜皮、西瓜籽像素级语义分割

TO DO LIST：

  1. Labelme 标注语义分割数据集（子豪兄已经帮你完成了）
  2. 划分训练集和测试集（子豪兄已经帮你完成了）
  3. Labelme 标注转 Mask 灰度图格式（子豪兄已经帮你完成了）
  4. 使用 MMSegmentation 算法库，撰写 config 配置文件，训练 PSPNet 语义分割算法
  5. 提交测试集评估指标
  6. 使用数据集之外的西瓜图片和视频进行预测，并存储展示预测的结果。
  7. 训练 Segformer 语义分割算法，提交测试集评估指标

西瓜瓤、西瓜籽数据集：

标注：同济子豪兄 image

image

image

数据集下载链接：

  • Labelme标注格式（没有划分训练集和测试集）：https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20230130-mmseg/dataset/watermelon/Watermelon87_Semantic_Seg_Labelme.zip
  • Mask标注格式（已划分训练集和测试集）：https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20230130-mmseg/dataset/watermelon/Watermelon87_Semantic_Seg_Mask.zip

需提交的测试集评估指标：（不能低于 baseline 指标的 50% ）

  • aAcc: 60.6200
  • mIoU: 21.1400
  • mAcc: 28.4600

需要提交的其他文件

  1. 训练config
  2. 训练日志log和输出文件夹work_dirs
  3. 测试图片视频原图即推理后的图片视频
  4. 评估指标，config路径地址，log地址，图片视频地址放在提交文件夹的readme.md文件中

chg0901 avatar Jun 13 '23 08:06 chg0901

We recommend using English or English & Chinese for issues so that we could have broader discussion.

mm-assistant[bot] avatar Jun 13 '23 08:06 mm-assistant[bot]

作业链接：https://github.com/xixihic/openmmlabhomework/tree/master/MMSegmentation 两个网络：PSPNet和Segformer, 都出现了训练感觉还可以，但是换张图想效果不是很好

xixihic avatar Jun 16 '23 06:06 xixihic

Github Link

https://github.com/chg0901/openmmlab2-hong/tree/main/Assignment4

作业结果

1. Config

./mmsegmentation/pspnet-Watermelon87_Semantic_Seg_Mask_20230616.py

2. Evaluation

  • aAcc: 94.2700
  • mIoU: 77.2600
  • mAcc: 86.7700

3. Training logs

./work_dirs/Watermelon87_Semantic_Seg_Mask/20230616_114954/20230616_114954.log

4. Test logs

Outputs is the same as the above train outputs since we test evey 400 epoch during training. .work_dirs/Watermelon87_Semantic_Seg_Mask/20230616_160732/20230616_160732.log

+------------+-------+-------+
|   Class    |  IoU  |  Acc  |
+------------+-------+-------+
|    Red     | 92.13 | 96.03 |
|   Green    | 87.87 | 91.65 |
|   White    | 77.85 | 91.88 |
| Seed-black | 84.83 | 93.94 |
| Seed-white | 68.25 | 73.52 |
| Unlabeled  | 52.67 |  73.6 |
+------------+-------+-------+
2023/06/16 12:09:59 - mmengine - INFO - Iter(val) [11/11]  
aAcc: 94.2700  mIoU: 77.2600  mAcc: 86.7700  data_time: 0.0016  time: 0.2658

2023/06/16 12:09:59 - mmengine - INFO - The previous best checkpoint /home/cine/Documents/GitHub/mmsegmentation/work_dirs/Watermelon87_Semantic_Seg_Mask/best_mAcc_iter_3600.pth is removed
2023/06/16 12:10:00 - mmengine - INFO - The best checkpoint with 86.7700 mAcc at 7600 iter is saved to best_mAcc_iter_7600.pth.
2023/06/16 12:10:00 - mmengine - INFO - The previous best checkpoint /home/cine/Documents/GitHub/mmsegmentation/work_dirs/Watermelon87_Semantic_Seg_Mask/best_mIoU_iter_2800.pth is removed
2023/06/16 12:10:01 - mmengine - INFO - The best checkpoint with 77.2600 mIoU at 7600 iter is saved to best_mIoU_iter_7600.pth.

5. Test and Visualization

Datafolder: ./mmsegmentation/data/

  1. Images

Image 1： https://github.com/chg0901/openmmlab2-hong/tree/main/Assignment4/mmsegmentation/data/watermelon_2.png Prediction1： https://github.com/chg0901/openmmlab2-hong/tree/main/Assignment4/mmsegmentation/data/watermelon_2_pred.png

Image 2： https://github.com/chg0901/openmmlab2-hong/tree/main/Assignment4/mmsegmentation/data/watermelon.png Prediction2： https://github.com/chg0901/openmmlab2-hong/tree/main/Assignment4/mmsegmentation/data/watermelon_pred.png

  1. Videos

original https://github.com/chg0901/openmmlab2-hong/tree/main/Assignment4/mmsegmentation/data/watermelon2.gif Prediction https://github.com/chg0901/openmmlab2-hong/tree/main/Assignment4/mmsegmentation/data/watermelon_pred2_.gif

chg0901 avatar Jun 17 '23 05:06 chg0901

代码仓地址：https://github.com/JeffDing/mmlabcamp/tree/main/第二期/mmsegmentation/Homework4.ipynb

训练config：在mmsegmentation/logs文件夹下 训练日志log和输出文件夹work_dirs：work_dirs太大了就不上传了 测试图片视频原图即推理后的图片视频：mmsegmentation/results目录下为结果文件 评估指标，提示没有注册数据集，还在研究中

JeffDing avatar Jun 17 '23 07:06 JeffDing

代码仓库地址：link

+------------+-------+-------+
|   Class    |  IoU  |  Acc  |
+------------+-------+-------+
| background |  91.0 | 92.18 |
|    red     |  93.7 | 97.95 |
|   green    | 60.05 |  92.1 |
|   white    | 81.88 | 90.53 |
| seed-black | 74.24 | 85.98 |
| seed-white | 57.59 | 71.57 |
+------------+-------+-------+
06/18 16:46:44 - mmengine - INFO - Iter(test) [11/11]    aAcc: 93.5200  mIoU: 76.4100  mAcc: 88.3800  data_time: 0.2213  time: 7.7713

Caczhtus avatar Jun 18 '23 09:06 Caczhtus

https://github.com/L-369-maha/OpenMMLab/tree/main/mmsegmentation

L-369-maha avatar Jun 19 '23 06:06 L-369-maha