Spaces:
Runtime error
Runtime error
Update README.md
Browse files
README.md
CHANGED
@@ -10,4 +10,34 @@ pinned: false
|
|
10 |
license: mit
|
11 |
---
|
12 |
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
license: mit
|
11 |
---
|
12 |
|
13 |
+
# Stamp Detection using YoloV9
|
14 |
+
|
15 |
+
This repository shows how to custom train your model using YOLOV9.
|
16 |
+
|
17 |
+
The script above uses a football players dataset to implement the process of custom training due to confidentiality with respect to the stamp dataset (PII information).
|
18 |
+
However the same script can be used traing using custom dataset.
|
19 |
+
|
20 |
+
## Steps
|
21 |
+
|
22 |
+
1.) Firstly we will install labelme using python pip. Labelme is a graphical image annotation tool
|
23 |
+
* pip install labelme
|
24 |
+
|
25 |
+
2.) Load your image directory in labelme and get to annotating
|
26 |
+
|
27 |
+
3.) Now once annotation is complete, install labelme2yolo to convert your labelme annotation to yolov9 format
|
28 |
+
* labelme2yolo --json_dir /path/to/labelme_json_dir/ --val_size 0.15 --test_size 0.15
|
29 |
+
|
30 |
+
4.) Once done you can execute above script by replacing dataset with your custom path
|
31 |
+
|
32 |
+
## Training Performance
|
33 |
+
The model was trained for 24 epochs on the stamp dataset. The training logs show the following metrics:
|
34 |
+
<img width="1161" alt="Screenshot 2024-05-17 at 8 56 30 PM" src="https://github.com/Himank-J/ERAV2/assets/55919214/dac6c631-fcf1-4931-a9d1-11e6f70a31b9">
|
35 |
+
|
36 |
+
As you can see from above training logs:
|
37 |
+
* Precision is 0.85
|
38 |
+
* Recall is 0.67
|
39 |
+
* mAp is 0.76
|
40 |
+
|
41 |
+
<img width="1013" alt="Screenshot 2024-05-12 at 1 49 36 PM" src="https://github.com/Himank-J/ERAV2/assets/55919214/f44a98f2-77b8-4168-a52b-25292afec069">
|
42 |
+
|
43 |
+
The above model is also available on huggingface for testing - https://huggingface.co/spaces/HimankJ/StampDetection
|