|
--- |
|
tags: |
|
- object-detection |
|
--- |
|
|
|
## Dataset |
|
|
|
The dataset was referenced in the Smartathon competition.It's consist of 7874 images annontated with 11 classes: |
|
|
|
* GARBAGE 8597 |
|
* CONSTRUCTION_ROAD 2730 |
|
* POTHOLES 2625 |
|
* CLUTTER_SIDEWALK 2253 |
|
* BAD_BILLBOARD 1555 |
|
* GRAFFITI 1124 |
|
* SAND_ON_ROAD 748 |
|
* UNKEPT_FACADE 127 |
|
* FADED_SIGNAGE 107 |
|
* BROKEN_SIGNAGE 83 |
|
* BAD_STREETLIGHT 1 |
|
|
|
The dataset highly imbalanced and contain some humman errors. |
|
|
|
## Our SEE Team Solution |
|
|
|
1. Convert from Pascal VOC to YOLO format |
|
2. Model Hyperparamter tuning |
|
3. Train the data on Yolov7 |
|
4. Evaluate the model |
|
5. Expalin Different techniques to Automation of Data Annotation |
|
|
|
For our solution detials: [notebook](https://colab.research.google.com/drive/1mo3HxJrg8wDGp_FhkB_0qAs41XvQ3hjR?usp=sharing) |
|
|
|
### How to use |
|
|
|
1. You can just download file weights from the files section |
|
2. clone yolov7 repo `!git clone https://github.com/WongKinYiu/yolov7` |
|
3. ensure your current working directory is yolov7 then run `! pip install -r requirements.txt` |
|
4. then run the detector script `! python detect.py --weights " model.pt path" --img 736 --conf 0.27 --source "testing image path" --save-txt` |
