Hi ARC Team,
Thank you for your great work on this project! While going through the code, I noticed a potential inconsistency in the vlm_response_mask function when handling the 'global' category.
When the category parameter is set to 'global', the mask is currently initialized as:
mask = 255 * np.zeros((height, width))
For global category processing, the mask should presumably cover the entire image area. This would be better implemented as:
mask = np.ones((height, width)) * 255
Would you kindly confirm if this is indeed an implementation oversight? Looking forward to your insights!
Hi ARC Team,
Thank you for your great work on this project! While going through the code, I noticed a potential inconsistency in the vlm_response_mask function when handling the 'global' category.
When the category parameter is set to 'global', the mask is currently initialized as:
mask = 255 * np.zeros((height, width))
For global category processing, the mask should presumably cover the entire image area. This would be better implemented as:
mask = np.ones((height, width)) * 255
Would you kindly confirm if this is indeed an implementation oversight? Looking forward to your insights!