- Description
- Usage
2-1. Model Training
2-2. Detection Evaluation
2-3. Result Analysis - Contact
This is a repository for PyTorch implementation of YOLOv1 following the original paper (https://arxiv.org/abs/1506.02640). In addition, we provide model weights that trained on the VOC dataset, so you can quickly train YOLOv1 on your own dataset ! Just download the weight given below and move it into "./weights" directory. If you wanna train YOLOv1 on your dataset from the scratch, add "--scratch" in training command.
- Performance Table
| Model | Dataset | Train | Valid | Size (pixel) |
mAP (@0.5:0.95) |
mAP (@0.5) |
Params (M) |
FLOPs (B) |
|---|---|---|---|---|---|---|---|---|
| YOLOv1 (Paper:page_with_curl:) |
PASCAL-VOC | trainval2007+2012 | test2007 | 448 | not reported | 63.4 | not reported | 40.16 |
| YOLOv1 VGG16 (Paper:page_with_curl:) |
PASCAL-VOC | trainval2007+2012 | test2007 | 448 | not reported | 66.4 | not reported | not reported |
| YOLOv1 VGG16 (Our:star:) |
PASCAL-VOC | trainval2007+2012 | test2007 | 448 | 34.9 | 67.2 | 25.49 | 127.00 |
| YOLOv1 VGG16-BN (Our:star:) |
PASCAL-VOC | trainval2007+2012 | test2007 | 448 | 36.9 | 69.4 | 25.49 | 127.43 |
| YOLOv1 Resnet18 (Our:star:) |
PASCAL-VOC | trainval2007+2012 | test2007 | 448 | 38.8 | 68.6 | 21.95 | 18.81 |
| YOLOv1 Resnet34 (Our:star:) |
PASCAL-VOC | trainval2007+2012 | test2007 | 448 | 43.2 | 72.6 | 32.06 | 29.01 |
| YOLOv1 Resnet50 (Our:star:) |
PASCAL-VOC | trainval2007+2012 | test2007 | 448 | 43.0 | 73.5 | 35.06 | 37.58 |
-
Pretrained Model Weights Download
- You can train your own YOLOv1 model using various backbone architectures of ResNet18, ResNet34, ResNet50, ResNet101, VGG16, and VGG16-BN. If you wanna train YOLOv1 on your dataset from the scratch, add "--scratch" in training command like below.
python train.py --exp my_test
--data voc.yaml
--backbone {vgg16, vgg16-bn, resnet18, resnet34, resnet50, resnet101}
--scratch(optional)- It computes detection metric via mean Average Precision(mAP) with IoU of 0.5, 0.75, 0.5:0.95. I follow the evaluation code with the reference on https://github.com/rafaelpadilla/Object-Detection-Metrics
python val.py --exp my_test --data voc.yaml --ckpt-name best.pt- After training is done, you will get the results shown below.
2022-11-25 18:37:35 | YOLOv1 Architecture Info - Params(M): 35.07, FLOPS(B): 32.41
2022-11-25 18:41:48 | [Train-Epoch:001] multipart: 13.6807 obj: 0.3439 noobj: 12.8928 box: 0.5445 cls: 4.1677
2022-11-25 18:45:52 | [Train-Epoch:002] multipart: 3.8190 obj: 0.4812 noobj: 0.1155 box: 0.3377 cls: 1.5916
2022-11-25 18:49:58 | [Train-Epoch:003] multipart: 3.3824 obj: 0.4848 noobj: 0.1571 box: 0.2936 cls: 1.3509
2022-11-25 18:54:05 | [Train-Epoch:004] multipart: 3.1404 obj: 0.4771 noobj: 0.1755 box: 0.2745 cls: 1.2028
2022-11-25 18:58:11 | [Train-Epoch:005] multipart: 3.0149 obj: 0.4663 noobj: 0.1998 box: 0.2640 cls: 1.1287
2022-11-25 19:02:17 | [Train-Epoch:006] multipart: 2.8718 obj: 0.4488 noobj: 0.2169 box: 0.2517 cls: 1.0560
2022-11-25 19:06:23 | [Train-Epoch:007] multipart: 2.7623 obj: 0.4314 noobj: 0.2359 box: 0.2440 cls: 0.9928
2022-11-25 19:10:29 | [Train-Epoch:008] multipart: 2.6833 obj: 0.4180 noobj: 0.2470 box: 0.2365 cls: 0.9595
2022-11-25 19:14:35 | [Train-Epoch:009] multipart: 2.6262 obj: 0.4060 noobj: 0.2590 box: 0.2335 cls: 0.9235
2022-11-25 19:18:43 | [Train-Epoch:010] multipart: 2.5375 obj: 0.3966 noobj: 0.2653 box: 0.2251 cls: 0.8827
2022-11-25 19:19:40 |
- Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.240
- Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.534
- Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.172
- Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.007
- Average Precision (AP) @[ IoU=0.50 | area= small | maxDets=100 ] = 0.031
- Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.087
- Average Precision (AP) @[ IoU=0.50 | area=medium | maxDets=100 ] = 0.249
- Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.298
- Average Precision (AP) @[ IoU=0.50 | area= large | maxDets=100 ] = 0.626
...
2022-11-26 05:07:31 | [Train-Epoch:149] multipart: 1.2090 obj: 0.2616 noobj: 0.2845 box: 0.1177 cls: 0.2167
2022-11-26 05:07:32 | [Best mAP at 140]
- Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.438
- Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.730
- Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.439
- Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.083
- Average Precision (AP) @[ IoU=0.50 | area= small | maxDets=100 ] = 0.191
- Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.199
- Average Precision (AP) @[ IoU=0.50 | area=medium | maxDets=100 ] = 0.445
- Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.510
- Average Precision (AP) @[ IoU=0.50 | area= large | maxDets=100 ] = 0.788
- Author: Jiho Park
- Email: pjh5672.dev@gmail.com