Yolo_Nano-master
所属分类：人工智能/神经网络/深度学习
开发工具：Python
文件大小：15431KB
下载次数：4
上传日期：2020-08-28 09:31:24
上 传 者：
xiaoxiao_wei
说明： 基于深度学习目标检测yolov3算法的源码
(Source code of yolov3 algorithm based on deep learning target detection)
文件列表:
__pycache__ (0, 2020-01-13)
__pycache__\test.cpython-36.pyc (4598, 2020-01-13)
assets (0, 2020-01-13)
assets\show.jpg (395394, 2020-01-13)
assets\structure.png (373149, 2020-01-13)
config (0, 2020-01-13)
config\MOT17-07.data (91, 2020-01-13)
config\coco.data (115, 2020-01-13)
config\coco_person.data (121, 2020-01-13)
config\create_custom_model.sh (8532, 2020-01-13)
config\custom.data (99, 2020-01-13)
config\multi_person.data (91, 2020-01-13)
config\vid.data (87, 2020-01-13)
config\vid_VID20191016164708.data (87, 2020-01-13)
config\yolo-nano.cfg (38, 2020-01-13)
config\yolo-nano_person.cfg (37, 2020-01-13)
config\yolov3-tiny.cfg (2025, 2020-01-13)
config\yolov3.cfg (8338, 2020-01-13)
config\yolov3_person.cfg (8332, 2020-01-13)
data (0, 2020-01-13)
data\MOT (0, 2020-01-13)
data\MOT\MOT17-07-DPM.txt (5500, 2020-01-13)
data\MOT\multi_person.txt (4950, 2020-01-13)
data\__pycache__ (0, 2020-01-13)
data\__pycache__\default_path.cpython-36.pyc (1200, 2020-01-13)
data\coco.names (625, 2020-01-13)
data\coco (0, 2020-01-13)
data\coco\5k.txt (85000, 2020-01-13)
data\coco\trainvalno5k.txt (2010879, 2020-01-13)
data\coco_person.names (7, 2020-01-13)
data\custom (0, 2020-01-13)
data\custom\classes.names (6, 2020-01-13)
data\custom\images (0, 2020-01-13)
data\custom\images\train.jpg (113300, 2020-01-13)
data\custom\labels (0, 2020-01-13)
data\custom\labels\train.txt (34, 2020-01-13)
data\custom\train.txt (29, 2020-01-13)
data\custom\valid.txt (29, 2020-01-13)
... ...
# PyTorch-YOLO_Nano
A minimal PyTorch implementation of YOLO_Nano
- [Yolo_Nano](https://arxiv.org/abs/1910.01271)
- [Bag of Freebies for Training Object Detection Neural Networks](https://arxiv.org/abs/1902.04103v3)
#### Trick In here I have done
[Bag of Freebies for Training Object Detection Neural Networks](https://arxiv.org/abs/1902.04103v3) tell us that fixup in object detection can increase the mAP, So I realize it and test in result.
- [x] Data Augmentation
- [x] Fixup
- [x] Cosine lr decay
- [x] Warm up
- [ ] multi-GPU
#### Download COCO
$ cd data/
$ bash get_coco_dataset.sh
## Module Pipeline
## training
```bash
bash train.sh
Better Para:
--epochs 120
--batch_size 8
--model_def ./config/yolo-nano_person.cfg
--lr 2.5e-4
--fix_up True
--lr_policy cosine
```
## Testing
```bash
python test.py --data_config ./config/coco_person.data --model_def ./config/yolo-nano_person.cfg --weights_path [checkpoint path]
```
## Result
In this engineer we only train our model using coco-train person class
we compare with yolov-3yolo-tiny. We got competitive results.
Methods |mAP@50|mAP|weights|FPS| Model
:--------------:|:--:|:--:|:--: |:--: |:--:
yolov3(paper) | 74.4 |40.3 | 204.8M| 28.6FPS |[Google Disk](https://pjreddie.com/media/files/yolov3.weights)
yolov3-tiny(paper) | 38.8 |15.6 | 35.4M | 45FPS |[Google Disk](https://pjreddie.com/media/files/yolov3-tiny.weights)
yolo-nano | 55.6 |27.7 | 22.0M | 40FPS |[Baidu WebDisk](https://pan.baidu.com/s/1Rp0is2LqA91XwjRc41mGaw)
Baidu WebDisk Key: p2j3
## Ablation Result
Augmentation| fixup | mAP
:--------------:|:--:|:--:
No|No|54.3
Yes|No|53.9
No|YES|55.6
YES|YES|54.8
## Inference Result
近期下载者：
相关文件：
收藏者：