...
- It is over files 1,2
- used denoise-log
- adding file 4, with subbkg, reduced acc to 63%
- adding file 4, without subbkg reduced acc to 48%
Third Pass
Here we did better de-noising and compared to a signal processing approach. The de-noising is:
- opencv medianBlur
- yag: 5pt
- vcc: 7pt
- opencv guassianBlur
- yag: 55 x 55
- vcc: 15 x 15
- lanczos reduction
After doing the de-noising, and before the reduction, we find the maximum value in the image and call it a hit if it is in the labeled box. This signal processing solution performs quite well. Over files 1,2,4 and doing the background subtraction for file 4, it does:
- 100% for the yag
- 99% for the vcc
- one of the vcc boxes is mislabeled though
- the other one, it is close to the box, the gaussian blur took a longer shape with some nearby noise and made it more round (we guess)
The regression pipeline does quite well on the yag, but less well on the vcc
yag: inter/union accuracies: th=0.50 acc=0.89 th=0.20 acc=0.97 th=0.01 acc=0.98
vcc: inter/union accuracies: th=0.50 acc=0.09 th=0.20 acc=0.38 th=0.01 acc=0.66
Here are plots
vcc median+Gaussian Blur
yag median+Guassian Blur