Open lihanlin99 opened 3 years ago
Hi. I am not sure whether this is still open or not, but I implemented a CrowdHuman (CH) ingestion script myself. It is, I believe, fairly straightforward to use. It is a command-line tool that takes the CH annotation file and produces a COCO-like annotation JSON file out of it. Subsequently, it can be easily loaded into the SiamMOT pipeline for training. It will behave as an image-based dataset, thus, all the necessary augmentations to fake the tracking process will seamlessly work. If there are any questions or suggestions, feel free to contact me.
Could you please release the CrowdHuman_val_fbox.json and CrowdHuman_val_vbox.json?