README.md

NigDet

Nighttime vehicle detection.
https://github.com/mozpp/NigDet

Test

1, download dataset and model. 链接: https://pan.baidu.com/s/17AEdluJq0hOByEHuc6Hnsg 提取码: 6dqp

2, ./mscnn-master/examples/kitti_car/run_mscnn_detection.m.

Train your own dataset

1, tools/blob1.m 获取候选灯光.

2, tools/blob_bbox.m: generate a txt file 'blob_stats.txt', to store feature and label for each blob_candi.

3, tools/crop_blob.m: crop each blob_candi according to 'blob_stats.txt', and store.

4, cd ./classify_pytorch to train a classifier to classify blob.

4.1 run main.py to train.
4.2 run test.py to test and store result 'blob_crop_result.txt'.

5, save vehicle highlight mask.

5.1 run tools/build_BlobDataset.m or tools/build_BlobDataset_Ver2.m to save classify result to txt 'blob_statsVer1.txt'.
5.2 run tools/settle_jan30.m or tools/settle_.m to store vehicle highlight mask.

6, tools/concat4ch.py concat mask to original image.

7, follow mscnn-master to compile caffe and generate training label file.
https://github.com/mozpp/mscnn-master

8, cd ./mscnn-master/examples/kitti_car/mscnn-p123s2-subcls3, run trian_mscnn.sh to train a model.