G-SAM2 for SKU-110K like data

[venu-banaras]

Hi,

I am trying to use this model for detecting SKUs in a dataset that highly resembles the SKU-110K dataset. The products are placed tightly to each other and their size vary vastly. I am able to detect large sized SKU like shampoo bottles and liquid detergent refill packets. But there is absolutely zero detection for small packets like shaving cream, razors, etc.

I am using this prompt
TEXT_PROMPT = "bottles. packets. single packets. cans. tubes. single items. products."

Using this, I am getting the above-said results. I know that using SAM2 as a standalone is able to detect everything, but the problem lies in IG Bert. I am also cropping the images shelf-wise to get the desired results.

Can anyone suggest how I can get accurate detection for said case? I am providing sample images, both whole and cropped, that I am working on.