MULTIPLE OBJECT in the images are not getting detected

[sumanttyagi]

trafficstars

what i have observed while running the pre trained model making detection . I was not able to make multiple detection say i want to detect a car but it will only detect one car not all the cars ?

• [ ] a) is this expected behavior one box per class ?
• [ ] b) how can i decode to get multiple bounding box for the same labels ?

[junchen14]

if you use the [detection] identifier, it is expected to extract all the bboxs for the car label.

which pretrained checkpoint are you loading? is it "MiniGPT-v2 (online developing demo)"? This checkpoint should give you more expected output.

[sumanttyagi]

Yes i am using MiniGPT-v2 (online developing demo).

while doing this i observed that in VQA it can answer that there are 3 cars in image, but while making detection it is only showing 1 . I wonder how it can predict 3 in vqa but not able to detect all the cars .

[junchen14]

Sometimes the model is sensitive to the singular and plural forms of inputs, such as 'car' and 'cars'. Maybe you can also explore different prompts.