Faster-RCNN and YOLOv3 for Object Detection

Used Framework

The project is built on top of the mmdetection.

How to train?

1. Prepare the environment: Users have to install the mmdetection framework and its dependencies. The installation guide can be found here. The project is built-from-source following the installation guide. After installation, cd into the mmdetection directory. bash cd mmdetection

2. Prepare the dataset: Download the Pascal VOC dataset using the following command: bash python tools/misc/download_dataset.py --dataset-name voc2007 However, we need to transform the dataset into the COCO format. To do this, run the following command: bash python voc_to_coco.py This script will create 3 annotation files in the VOCdevkit directory: voc07_train.json, voc07_val.json, and voc07_test.json. Then, we need to utilize the script split.py where you need to modify line4 and line6 in order to seperate images into training and validation and test sets. After that, you can create a new folder named coco inside data directory and move the voc07_train.json, voc07_val.json, and voc07_test.json files into the coco directory. Remember to rename the files to instances_train2017.json, instances_val2017.json, and instances_test2017.json respectively. After that, the data/coco directory should look like this: data/coco annotations instances_train2017.json instances_val2017.json instances_test2017.json test2017 train2017 val2017

3. Train the model: To train the model, run the following command: bash python tools/train.py configs/faster_rcnn/faster-rcnn_r50_fpn_1x_coco.py --work-dir ckpt/faster_rcnn

or bash python tools/train.py configs/yolov3/yolov3_d53_8xb8-ms-608-273e_coco.py --work-dir ckpt/yolov3 The training process will start and the model will be saved in the work_dirs directory. One can visualize the loss using tensorboard. 4. Test the model: To test the model, run the following command: bash python tools/test.py configs/faster_rcnn/faster-rcnn_r50_fpn_1x_coco.py ckpt/faster_rcnn/latest.pth --show-dir results/faster_rcnn

or bash python tools/test.py configs/yolov3/yolov3_d53_8xb8-ms-608-273e_coco.py ckpt/yolov3/latest.pth --show-dir results/yolov3 The testing process will start and the results will be saved in the results directory.

1. Inference: To perform inference on an image, run the following command: bash python demo/image_demo.py path/to/your/image configs/faster_rcnn/faster-rcnn_r50_fpn_1x_coco.py ckpt/faster_rcnn/latest.pth --out-dir inference_results

How to visualize the region proposals in the first stage of Faster-RCNN?

Run the command: bash python ./tools/test.py configs/rpn/rpn_r50_fpn_1x_coco.py path/to/your/ckpt --show-dir faster_rcnn_output2/rpn_proposal

Pretrained model

Checkpoints are available at : https://pan.baidu.com/s/1CSwTuAoM4eOzkBDr2FY2NA : tbjb --v6.