Even though my predictions are correct when it was visually analysed by drawing bounding boxes using cv2 rectangle. I'm getting 0 as my score when the results are uploaded. Please help me with this issue
Posted by: hrishikeshps @ March 9, 2021, 3:35 p.m.1. You mean you manually created bounding boxes for dry-run data and submitted?
2. Have you compared your submission to the example submission file?
1. The predictions that I got based on a detection algorithm was used to draw rectangles on dry run samples. On visual analysis of the predicted bounding boxes, the predicted boxes had reasonable accuracy.
2. Yes, I checked the same with the format of sample result submission text files and the format looked similar.
Is the format of the text file (class name, x1,y1,x2,y2,confidence), where(x1,y1) is the top left corner points and (x2,y2) is bottom right corner points?
1. Could you double-check the class names (car, bus, bicycle, motorcycle, pedestrian) and also the value range (for x1 y1 x2 y2 confidence) are reasonable and comparable to the example?
2. Note that the example submission gets very low mAP, you probably want to get different results from the example.
Yes the format is correct, (class name, x1,y1,x2,y2,confidence)
Posted by: shakehealthy @ March 9, 2021, 6:09 p.m.I even compared my results with a public submission and the results were very comparable. The class names which I initially used were similar to what you have said but on further inspection I changed the class names from pedestrian to person and motorcycle to motorbike.
Posted by: hrishikeshps @ March 9, 2021, 6:15 p.m.How about you send your latest submission to cvpr2021.ug2challenge@gmail.com?
We may take some time to look into that.
Posted by: shakehealthy @ March 9, 2021, 6:32 p.m.Ok, Thank you
Posted by: hrishikeshps @ March 9, 2021, 6:33 p.m.