Here is my pytorch implementation of the model described in the paper YOLO9000: Better, Faster, Stronger paper.
An example of my model's output.
With my code, you can:
I used 4 different datases: VOC2007, VOC2012, COCO2014 and COCO2017. Statistics of datasets I used for experiments is shown below
Dataset | Classes | #Train images/objects | #Validation images/objects |
---|---|---|---|
VOC2007 | 20 | 5011/12608 | 4952/- |
VOC2012 | 20 | 5717/13609 | 5823/13841 |
COCO2014 | 80 | 83k/- | 41k/- |
COCO2017 | 80 | 118k/- | 5k/- |
Create a data folder under the repository,
cd {repo_root}
mkdir data
VOC: Download the voc images and annotations from VOC2007 or VOC2012. Make sure to put the files as the following structure:
VOCDevkit
├── VOC2007
│ ├── Annotations
│ ├── ImageSets
│ ├── JPEGImages
│ └── ...
└── VOC2012
├── Annotations
├── ImageSets
├── JPEGImages
└── ...
COCO: Download the coco images and annotations from coco website. Make sure to put the files as the following structure:
COCO
├── annotations
│ ├── instances_train2014.json
│ ├── instances_train2017.json
│ ├── instances_val2014.json
│ └── instances_val2017.json
│── images
│ ├── train2014
│ ├── train2017
│ ├── val2014
│ └── val2017
└── anno_pickle
├── COCO_train2014.pkl
├── COCO_val2014.pkl
├── COCO_train2017.pkl
└── COCO_val2017.pkl
Epoches | Learning rate |
---|---|
0-4 | 1e-5 |
5-79 | 1e-4 |
80-109 | 1e-5 |
110-end | 1e-6 |
You could find all trained models I have trained in YOLO trained models
For each dataset, I provide 2 different pre-trained models, which I trained with corresresponding dataset:
You could specify which trained model file you want to use, by the parameter pre_trained_model_type. The parameter pre_trained_model_path then is the path to that file.
If you want to train a model with a VOC dataset, you could run:
If you want to train a model with a COCO dataset, you could run:
If you want to train a model with both COCO datasets (training set = train2014 + val2014 + train2017, val set = val2017), you could run:
For each type of dataset (VOC or COCO), I provide 3 different test scripts:
If you want to test a trained model with a standard VOC dataset, you could run:
If you want to test a model with some images, you could put them into the same folder, whose path is path/to/input/folder, then run:
If you want to test a model with a video, you could run :
I trained models in 2 machines, one with NVIDIA TITAN X 12gb GPU and the other with NVIDIA quadro 6000 24gb GPU.
The training/test loss curves for each experiment are shown below:
Statistics for mAP will be updated soon ...
Some output predictions for experiments for each dataset are shown below: