Issues in Shape Handling and Original Image Size

[gongshichina]

@hanoonaR Thank you for your open-source contribution! I have noticed two potential bugs in the code. Could you please take a look and verify them?

Isuse1: https://github.com/mbzuai-oryx/groundingLMM/blob/4073365f652f9ea27ee53daabb417cb4da8361de/app.py#L227

The global_enc_image shape is (1, 3, 336, 336). It seems that global_enc_image.shape[2:4] should be used instead.

Issue 2: https://github.com/mbzuai-oryx/groundingLMM/blob/4073365f652f9ea27ee53daabb417cb4da8361de/model/layers.py#L302

The size of the original image appears to be 336 rather than 224. This mismatch might lead to unexpected behavior in the code.