[Question] Use of tiling mechanism for SSD models

[parthjdoshi]

Hey @GeekAlexis , I had a query regarding the use of tiles in the SSD models. Are you doing it to create batches and improve inference time? Would it be considerably slower or inaccurate to just pass the whole image, rather than passing individual tiles?

[GeekAlexis]

@arsenal-2004 Yes, it would have more detection misses without tiling because the pretrained SSD models come with small input size (300x300). You can try it out by using “tiling_grid": [1, 1] in mot.json

[parthjdoshi]

@GeekAlexis I understand. Thank you