replace img_path to media_subpath

mill413 · Apr 21, 2024 · fca1dcd · fca1dcd
1 parent fa6eefa
commit fca1dcd
Show file tree

Hide file tree

Showing 68 changed files with 68 additions and 68 deletions.
diff --git a/_posts/journal/2023-10-12-day1.md b/_posts/journal/2023-10-12-day1.md
@@ -4,7 +4,7 @@ title: Day1
 date: 2023-10-12 17:51 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 今天啥也没干，给虚拟机里的系统全更新了一遍，明天一定要学习！

diff --git a/_posts/journal/2023-10-13-day2.md b/_posts/journal/2023-10-13-day2.md
@@ -4,7 +4,7 @@ title: Day2
 date: 2023-10-13 21:42 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 找了个教程跑个代码：<https://blog.csdn.net/darkredrock/article/details/131049498>

diff --git a/_posts/journal/2023-10-14-day3.md b/_posts/journal/2023-10-14-day3.md
@@ -4,7 +4,7 @@ title: Day3
 date: 2023-10-14 19:54 +0800
 category: [读研日记]
 tags: [label-studio]
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 今天好像又啥也没干，这种日记还有记录的必要吗

diff --git a/_posts/journal/2023-10-15-day4.md b/_posts/journal/2023-10-15-day4.md
@@ -4,7 +4,7 @@ title: Day4
 date: 2023-10-15 23:13 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 今天又是睡了一整个白天

diff --git a/_posts/journal/2023-10-21-day5.md b/_posts/journal/2023-10-21-day5.md
@@ -4,7 +4,7 @@ title: Day5
 date: 2023-10-21 22:33 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 好几天没学习了，今天打了几个标，一定要去实验室学习了

diff --git a/_posts/journal/2023-10-23-day6.md b/_posts/journal/2023-10-23-day6.md
@@ -4,7 +4,7 @@ title: Day6
 date: 2023-10-23 22:05 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 今天看了一下[这个项目的代码](https://github.com/Linaom1214/U-Transformer/tree/master)，对这个数据集的处理有了点启发，这个数据集里有很多目标在图上只有几个像素点，对于这些数据我可以采用先前的思路，直接以数据集里给的目标的坐标为中心，做一个3x3的标注框

diff --git a/_posts/journal/2023-10-24-day7.md b/_posts/journal/2023-10-24-day7.md
@@ -4,7 +4,7 @@ title: Day7
 date: 2023-10-24 23:03 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 今天没有读论文和做数据集，熟悉了一下yolov8的api，把集群上的文件结构改了一下，重要的东西拿出来写了个脚本，他竟然不允许指定save_dir！愤怒喵

diff --git a/_posts/journal/2023-11-17-day8.md b/_posts/journal/2023-11-17-day8.md
@@ -4,7 +4,7 @@ title: Day8
 date: 2023-11-17 22:01 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 好久没写日记了，到开题的日子了

diff --git a/_posts/journal/2023-11-28-day9.md b/_posts/journal/2023-11-28-day9.md
@@ -4,7 +4,7 @@ title: Day9 - 集群上的虚拟环境配置
 date: 2023-11-28 00:45 +0800
 category: [读研日记]
 tags: [conda,pytorch,ubuntu]
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 心血来潮把集群上的conda环境全删了重新弄，然后悲剧了，环境下不回来了

diff --git a/_posts/journal/2023-11-30-day10.md b/_posts/journal/2023-11-30-day10.md
@@ -4,7 +4,7 @@ title: Day10 - 两篇基于YOLO的背景差分改进论文
 date: 2023-11-30 22:56 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/231130/"
+media_subpath: "/assets/img/posts/231130/"
 ---
 
 今天看了两个加了背景差分法的yolo做目标检测的论文

diff --git a/_posts/journal/2023-12-04-day11.md b/_posts/journal/2023-12-04-day11.md
@@ -4,7 +4,7 @@ title: Day11
 date: 2023-12-04 18:28 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 尝试用`yolov8s`和`yolov5s`跑个baseline，脚本改了又改，300epochs跑出来过拟合了，跑一次200的试试

diff --git a/_posts/journal/2023-12-06-day12.md b/_posts/journal/2023-12-06-day12.md
@@ -4,7 +4,7 @@ title: Day12
 date: 2023-12-06 11:55 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/231206/"
+media_subpath: "/assets/img/posts/231206/"
 ---
 
 代码重构了一下，把ultralytics放了进来方便直接改源码，后面慢慢把不需要的部分删掉再传到github

diff --git a/_posts/journal/2023-12-07-day13.md b/_posts/journal/2023-12-07-day13.md
@@ -4,7 +4,7 @@ title: Day13 - 删除ultralytics/assets报错
 date: 2023-12-07 16:23 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 发现了ultralytics的一个bug

diff --git a/_posts/journal/2023-12-08-day14.md b/_posts/journal/2023-12-08-day14.md
@@ -4,7 +4,7 @@ title: Day14
 date: 2023-12-08 10:46 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 很高兴SDOA-D是正矩形框，写了个脚本处理成yolo格式，但是他是高分辨率下的小目标，难说

diff --git a/_posts/journal/2023-12-09-day15.md b/_posts/journal/2023-12-09-day15.md
@@ -4,7 +4,7 @@ title: Day15
 date: 2023-12-09 23:12 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 参照HIC-YOLO把yolov8-cbam里的cbam模块放到最后, 跑一遍yolov8m-cbam、yolov8m-p234、yolov5m-cbam

diff --git a/_posts/journal/2023-12-10-day16.md b/_posts/journal/2023-12-10-day16.md
@@ -4,7 +4,7 @@ title: Day16
 date: 2023-12-10 11:52 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 把新跑出来的数据合并一下，修改的yolov8-cbam效果还没有原来的好，就不合并了，把cbam模块放回到第一个c2f后面

diff --git a/_posts/journal/2023-12-11-day17.md b/_posts/journal/2023-12-11-day17.md
@@ -4,7 +4,7 @@ title: Day17
 date: 2023-12-11 12:06 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 不知道跑了个什么模型，模型名是yolov8-cb，只能从模型结构判断可能是yolov8-cbam

diff --git a/_posts/journal/2023-12-12-day18.md b/_posts/journal/2023-12-12-day18.md
@@ -4,7 +4,7 @@ title: Day18
 date: 2023-12-12 16:09 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov8s为基准，200epochs，在VisDrone2019上训练结果

diff --git a/_posts/journal/2023-12-13-day19.md b/_posts/journal/2023-12-13-day19.md
@@ -4,7 +4,7 @@ title: Day19
 date: 2023-12-13 11:56 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5s为基准，200epochs，在VisDrone2019上训练结果

diff --git a/_posts/journal/2023-12-14-day20.md b/_posts/journal/2023-12-14-day20.md
@@ -4,7 +4,7 @@ title: Day20 - yolov5s-tph在VisDrone下的表现
 date: 2023-12-14 22:31 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 昨天的acmix一直调不通代码，总是报`RuntimeError: Input type (float) and bias type (c10::Half) should be the same`，查了一下是因为model在验证的时候half了，参考<https://github.com/LeapLabTHU/ACmix/issues/18#issuecomment-1730733557>

diff --git a/_posts/journal/2023-12-15-day21.md b/_posts/journal/2023-12-15-day21.md
@@ -4,7 +4,7 @@ title: Day21
 date: 2023-12-15 21:59 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 整理代码的时候发现cbam一直用的hic里的CBAMBottleneck,改成了ultralytics提供的CBAM,map竟然涨了不少，太奇怪了

diff --git a/_posts/journal/2023-12-17-day22.md b/_posts/journal/2023-12-17-day22.md
@@ -4,7 +4,7 @@ title: Day22 - CBAM在yolov5s上不同添加位置在VisDrone上的对比实验
 date: 2023-12-17 20:19 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 对CBAM不同添加位置的结果

diff --git a/_posts/journal/2024-01-09-day23.md b/_posts/journal/2024-01-09-day23.md
@@ -4,7 +4,7 @@ title: Day23 - yolov8s和yolov5s的CBAM和P234改进在FLIR上的消融实验
 date: 2024-01-09 10:29 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 用FLIR跑了一遍之前的模型

diff --git a/_posts/journal/2024-01-14-day24.md b/_posts/journal/2024-01-14-day24.md
@@ -4,7 +4,7 @@ title: Day24
 date: 2024-01-14 12:15 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 在FLIR上跑yolov8的几个scale作为baseline， 再跑对应的cbam+p234作为对比

diff --git a/_posts/journal/2024-01-16-day25.md b/_posts/journal/2024-01-16-day25.md
@@ -4,7 +4,7 @@ title: Day25 - yolov8所有scale及对应CBAM+P234改进在FLIR上的对比实
 date: 2024-01-16 20:13 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 花了一周把yolov8的所有scale以及相应的cbam+p234的改进版都跑了一遍

diff --git a/_posts/journal/2024-01-18-day26.md b/_posts/journal/2024-01-18-day26.md
@@ -4,7 +4,7 @@ title: Day26
 date: 2024-01-18 10:18 +0800
 category: [读研日记]
 tags: [jupyter lab]
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 之前一直在学校的集群上用jupyterlab的环境，有段时间无法新建终端，报```/bin/zsh can`t open input file: cd ~```的错，昨天终于找到问题出在配置文件上

diff --git a/_posts/journal/2024-03-02-day27.md b/_posts/journal/2024-03-02-day27.md
@@ -4,7 +4,7 @@ title: Day27 - YOLO-TLA
 date: 2024-03-02 14:52 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/240302"
+media_subpath: "/assets/img/posts/240302"
 ---
 
 看了一篇YOLO-TLA

diff --git a/_posts/journal/2024-03-08-day28.md b/_posts/journal/2024-03-08-day28.md
@@ -4,7 +4,7 @@ title: Day28 - 两个新的小目标数据集
 date: 2024-03-08 14:37 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 目前正在使用的数据集主要是FLIR和VisDrone2019，还有AI-TOD和SODA-D两个数据集，但是几乎没用过

diff --git a/_posts/journal/2024-03-11-day29-baseline_visdrone.md b/_posts/journal/2024-03-11-day29-baseline_visdrone.md
@@ -4,7 +4,7 @@ title: Day29 - Baseline在VsiDrone上的表现
 date: 2024-03-11 23:33 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 以yolov5n/s/m和yolov8n/s作为baseline,在VisDrone2019上训练200epochs

diff --git a/_posts/journal/2024-03-12-day30-baseline_flir.md b/_posts/journal/2024-03-12-day30-baseline_flir.md
@@ -4,7 +4,7 @@ title: Day30 - Baseline在FLIR上的表现
 date: 2024-03-12 20:04 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 以yolov5n/s/m和yolov8n/s作为baseline,在FLIR上训练200epochs

diff --git a/_posts/journal/2024-03-14-day31-p2_flir.md b/_posts/journal/2024-03-14-day31-p2_flir.md
@@ -4,7 +4,7 @@ title: Day31 - P2改进在FLIR上的表现
 date: 2024-03-14 14:22 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅添加p2检测头,在FLIR上训练200epochs

diff --git a/_posts/journal/2024-03-16-day32-p2_visdrone.md b/_posts/journal/2024-03-16-day32-p2_visdrone.md
@@ -4,7 +4,7 @@ title: Day32 - P2改进在VisDrone上的表现
 date: 2024-03-16 12:07 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅添加p2检测头,在VisDrone上训练200epochs

diff --git a/_posts/journal/2024-03-16-day33-nwd_visdrone.md b/_posts/journal/2024-03-16-day33-nwd_visdrone.md
@@ -4,7 +4,7 @@ title: Day33 - NWD在VisDrone上的表现
 date: 2024-03-16 18:51 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅使用NWD损失函数,在VisDrone2019上训练200epochs

diff --git a/_posts/journal/2024-03-17-day34-nwd_flir.md b/_posts/journal/2024-03-17-day34-nwd_flir.md
@@ -4,7 +4,7 @@ title: Day34 - NWD在FLIR上的表现
 date: 2024-03-17 10:31 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅使用NWD损失函数,在FLIR上训练200epochs

diff --git a/_posts/journal/2024-03-18-day35-p234_flir.md b/_posts/journal/2024-03-18-day35-p234_flir.md
@@ -4,7 +4,7 @@ title: Day35 - 删除P5检测头+添加P2检测头（P234）在FLIR上的表现
 date: 2024-03-18 18:26 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s删除P5检测头+添加P2检测头,在FLIR上训练200epochs

diff --git a/_posts/journal/2024-03-19-day36-p234_visdrone.md b/_posts/journal/2024-03-19-day36-p234_visdrone.md
@@ -4,7 +4,7 @@ title: Day36 - 删除P5检测头+添加P2检测头（P234）在VisDrone上的表
 date: 2024-03-19 22:13 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s删除P5检测头+添加P2检测头,在VisDrone2019上训练200epochs

diff --git a/_posts/journal/2024-03-23-day37-cbam_flir.md b/_posts/journal/2024-03-23-day37-cbam_flir.md
@@ -4,7 +4,7 @@ title: Day37 - Backbone中C3/C2f后添加一层CBAM在FLIR上的表现
 date: 2024-03-23 19:31 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅在backbone的C3/C2f模块后添加一层CBAM,在FLIR上训练200epochs

diff --git a/_posts/journal/2024-03-26-day38.md b/_posts/journal/2024-03-26-day38.md
@@ -4,7 +4,7 @@ title: Day38 - Neck中C3/C2f后添加一层CBAM在FLIR上的表现
 date: 2024-03-26 12:05 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅在neck的C3/C2f模块后添加一层CBAM,在FLIR上训练200epochs

diff --git a/_posts/journal/2024-03-27-day39.md b/_posts/journal/2024-03-27-day39.md
@@ -4,7 +4,7 @@ title: Day39 - Neck中C3/C2f后添加一层CBAM在VisDrone上的表现
 date: 2024-03-27 23:02 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅在neck的C3/C2f模块后添加一层CBAM,在VisDrone2019上训练200epochs

diff --git a/_posts/journal/2024-03-28-day40-cbam_visdrone.md b/_posts/journal/2024-03-28-day40-cbam_visdrone.md
@@ -4,7 +4,7 @@ title: Day40 - Backbone中C3/C2f后添加一层CBAM在VisDrone上的表现
 date: 2024-03-28 15:42 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅在backbone的C3/C2f模块后添加一层CBAM,在VisDrone2019上训练200epochs

diff --git a/_posts/journal/2024-03-30-day41.md b/_posts/journal/2024-03-30-day41.md
@@ -4,7 +4,7 @@ title: Day41 - Neck中C3/C2f后添加一层GAM在VisDrone上的表现
 date: 2024-03-30 09:24 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅在neck的C3/C2f模块后添加一层GAM,在VisDrone2019上训练200epochs

diff --git a/_posts/journal/2024-04-01-day42.md b/_posts/journal/2024-04-01-day42.md
@@ -4,7 +4,7 @@ title: Day42 - Neck中C3/C2f后添加一层GAM在FLIR上的表现
 date: 2024-04-01 10:22 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅在neck的C3/C2f模块后添加一层GAM,在FLIR上训练200epochs

diff --git a/_posts/journal/2024-04-02-day43-gam_flir.md b/_posts/journal/2024-04-02-day43-gam_flir.md
@@ -4,7 +4,7 @@ title: Day43 - Backbone中C3/C2f后添加一层GAM在FLIR上的表现
 date: 2024-04-02 15:51 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅在backbone的C3/C2f模块后添加一层GAM,在FLIR上训练200epochs

diff --git a/_posts/journal/2024-04-04-day44-gam-visdrone.md b/_posts/journal/2024-04-04-day44-gam-visdrone.md
@@ -4,7 +4,7 @@ title: Day44 - Backbone中C3/C2f后添加一层GAM在VisDrone上的表现
 date: 2024-04-04 10:13 +0800
 category: [读研日记]
 tags: []
-img_path: "/assets/img/posts/"
+media_subpath: "/assets/img/posts/"
 ---
 
 yolov5n/s/m和yolov8n/s仅在backbone的C3/C2f模块后添加一层GAM,在VisDrone上训练200epochs