ssd's Introduction

This project aims to implement an SSD algorithm

Introduction

SSD is simply relative to methods that require object proposals because it eliminates proposal generation and subsequent pixel or feature resampling stages and encapsulates all computation in a single network

SSD discretizes the output space of bounding boxes into a set of default boxes over different aspect ratios and scales per future map location

Experiment Table

Backbone	Dataset	Training dataset	Valid dataset	Image size	mAP	mAP_50	mAP_75
Original paper	PASCAL VOC	trainval 2007+2012	test2007	300x300	--	74.3	--
Our implementation	PASCAL VOC	trainval 2007+2012	test2007	300x300	3.87	66.9	39.9

Dataset

Download Pascal VOC train+val 2012+2007
Download Pascal VOC test 2007 s Put all images, annotations, and .txt files in dataset/VOC folder as follows:

├── VOC
    ├── images
        ├── trainval2007
            ├── 000005.jpg
            ├── 000007.jpg
        ├── trainval2012
        ├── test2007
    ├── images_id
        ├── trainval2007.txt
        ├── trainval2012.txt
        ├── test2007.txt
    ├── labels
        ├── trainval2007
            ├── 000005.xml
            ├── 000007.xml
        ├── trainval2012
        ├── trainval2007

Training

python -m src.train --model_type vgg16/vgg16-bn --resume resume_most_recent_training

Evaluate

python -m src.eval --model_type vgg16/vgg16-bn --weight_type path_to_weight_best.pt

Prediction

python -m src.predict --model_type rvgg16/vgg16-bn--weight_path path_to_weight_best.pt --input_folder path_input_folder --output_folder path_to_output_folder --conf_thresh confidence_threshold --iou_thresh iou threshold