FingerRec / Faster-rcnn_Ship_detection

Ocean Ship Image Process(DSP) And Detection(faster-rcnn).

Requirements

Install these toolboxs from MatlabR2018b app manager directly.
A cuda-avilable gpu is auto used in matlab. Make sure you have a GPU, cpu train is very slow.

Screenshots

Introduction

This project include 3 parts.

• Preprocess
• Image Enhancement
• Ship Detection

Preprocess: denoise(medfilt), motionclear(winner filter), compress(wavelet compress) Enhancement: DCT, Color based Ship Detection: Faster-rcnn

Run

1. download MASTAI and airbus-ship-detection-data, extract to any place. (MASTAI only have main class label without mask. Kaggle airbus data are used to train faster-rcnn.)
2. add this project to path.
3. Modify line7-line9 in FasterRcnn/train_faster_rcnn.m according to your dataset's place, train faster-rcnn and use FasterRcnn/test_faster_rcnn.m to test the model.(Install resnet50 according to the tip.)
4. Test the model's preformance on MASTAI by using test_another_dataset.m.(modify line 5 and line 12.)
5. Open main.mlapp and run.

Trained Model

I have published trained model: mlp_based_faster_rcnn resnet50_based_faster_rcnn

Others

This project is based on matlab toolobox.