TDA2EVM5777: Adaboost object detection training tool

You need to provide both images and annotations.

1. The tool currently takes in .seq or .mp4 files for images. (It should be possible to modify the code to be able to specify a folder containing images instead of .seq or .mp4)

2. Annotation format supported is .vbb

You can fiund lots of .seq files (with the corresponding .vbb annotations) here. 

http://www.vision.caltech.edu/Image_Datasets/CaltechPedestrians/

An annotations/labelling code (matlab based) is also available in the above link.

An enhanced version of the same labeling tool (supporting annotation of .mp4 files) is also provided here:

https://github.com/tidsp/acf-jacinto/blob/master/vbb/vbbLabeler.m

Note - you don't have to annotate all the frames. acf-jacinto is intelligent to understand which are the annotated frames and use only those for training. This feature is especially useful when dealing with .mp4 videos.

After you do the training a .descriptor file will be written out. You can use this file in TDA2 VisionSDK Object Detection module. You can get more details in the documentation for the Object Detection module.

Best regards,