Open Masrur02 opened 3 weeks ago
Moreover, is there any way to increase the FPS? TIA
I have noticed that the confidence score is different when using the same model on these two images, would you like to check if the annotated frame is the same or not in your code.
Hi, Ya this is kind of my question. Why the confidence score is different and why the road is even not detected in the second image (though has a higher confidence). Moreover, when I tried the tracking with continuous id example the result was super good (you can see the ego car is also detected as the road, which is very weird, but in the tracking example it is perfectly predicted as a car). But in the tracking example it seems all the images from the video are stored first. Is there a way to extract images one by one and use the tracking example. Mainly I am planning to use the Grounded-SAM2 for tracking cars and roads from the images captured from a camera since your work seems very promising and interesting.
TIA
Hi, When I ran the code python grounded_sam2_local_demo.py the result was good with a prompt text="car. road."
But, when I have modified the code to read images from a video file and keep looping
the result has become very bad
What is the reason? Can you please help??
TIA