search

Zhongdao / Towards-Realtime-MOT

Joint Detection and Embedding for fast multi-object tracking
MIT License
2.37k stars 541 forks source link

Question: .seq in Caltech Pedestrian dataset #139

Closed anhnktp closed 4 years ago

anhnktp commented 4 years ago

Great works !

I followed instruction at: https://github.com/Zhongdao/Towards-Realtime-MOT/blob/master/DATASET_ZOO.md. However, with Caltech Pedestrian dataset, I downloaded all .tar file from: http://www.vision.caltech.edu/Image_Datasets/CaltechPedestrians/datasets/USA/ but I don't know how to convert .seq file to sequence of .png file (or .jpg) With City Person dataset, I downloaded from Gdrive link but it seems Citypersons.z03 file is not final part zip

Could you pls share link download Cityperson & Caltech_pedestrian dataset (have images & label_with_ids)

One more question, do your code support change custom detector instead of Yolov3 ? Thank you so much !

Zhongdao commented 4 years ago

Hi please check this tool https://github.com/mitmul/caltech-pedestrian-dataset-converter I will update the citypersons data and annotations, once done I will let you know. Currently, we have not provided a different detector, but if this is your need, you can modify the detector by yourself.

Zhongdao commented 4 years ago

@anhnktp I just checked the data in Gdrive and it seems okay. I think the final part zip is Cityperons.zip. Could you try this:

cat Citypersons.z01 Citypersons.z02 Citypersons.z03 Citypersons.zip > c.zip
unzip c.zip
anhnktp commented 4 years ago

@Zhongdao Many thanks !! I tried again and extracted succesfully Cityperson dataset. But, folder 'images' has test set (6 sequences) but folder 'labels_with_ids' don't have, is that right ? Tks. I converted Caltech dataset by that tool.

Zhongdao commented 4 years ago

@Zhongdao Yes, we do not use the test set, just leave them alone