search

Li-Chengyang / MSDS-RCNN

''Multispectral Pedestrian Detection via Simultaneous Detection and Segmentation''. BMVC 2018.
MIT License
64 stars 19 forks source link

Training Set size #5

Closed anushl9o5 closed 5 years ago

anushl9o5 commented 5 years ago

In the paper, there is a mention of 7,601 training images being used. But also has indicated following the method used in Illumination-aware faster r-cnn for robust multispectral pedestrian detection that uses 25,086 training images. Is there any further processing or filtering done to reduce them to 7,601 images ?

Li-Chengyang commented 5 years ago

Yes, we filtered the training data. You can refer to papers for details of our filtering method.

maartenvds commented 5 years ago

What about the labels 'cyclist' or 'people' (group of pedestrians)? If a training image contains one of them, do you include the image for training or not? (not mentioned in the paper)