Results (FPGA)

Final Results

Rank Team Name Precision Recall F1-Score FPS Total Score
1 SEUer 0.599 0.434 0.504 480.67 122.098
2 InvolutionNet 0.788 0.483 0.599 236.30 84.785
3 FSBIN 0.661 0.421 0.514 232.59 61.449
4 Puff 0.666 0.405 0.504 232.96 59.176
5 XDD 0.661 0.421 0.514 185.57 49.027
6 PCCC 0.661 0.421 0.514 138.60 36.618
7 approxitrack 0.539 0.245 0.337 164.71 18.706
8 Saecheonnyeon 0.740 0.432 0.545 43.09 12.799
9 bitsplicer 0.609 0.657 0.632 22.67 9.055
10 ADARLAB 0.804 0.601 0.688 7.25 3.432
11 fpgaconvnet 0.611 0.407 0.488 10.86 2.586
12 husky 0.743 0.465 0.572 4.97 1.626

Prelim #3

Rank Team Name Precision Recall F1-Score FPS Total Score
1 InvolutionNet 0.716 0.563 0.63 51.36 20.384784
2 ADARLAB 0.806 0.591 0.682 7.61 3.539594
3 bitsplicer 0.653 0.672 0.662 7.82 3.427068
4 Puff 0.625 0.471 0.537 2.29 0.660365
5 Log_gehts 0.01 0.013 0.011 3.39 0.00041
6 saecheonnyeon 0.004 0.002 0.002 15.63 0.000063
7 x2_team 0.004 0.002 0.003 3.49 0.000031

Did not run successfully:

  • FSBIN: ValueError: could not broadcast input array from shape (544,960,3) into shape (384,640,3) on line: image_input[0,...] = input
  • HLSPoliTo: IndexError: list index out of range on line: pred = non_max_suppression_custom(pred)[0])
  • NewLab: Board crashes on first batch. I reduced the batch size to 108, but now get an error UnboundLocalError: local variable 'i' referenced before assignment on line last_name = [k[0].name for k in rgb_imgs[i : BATCH_SIZE + i]]

Prelim #2

Note: These results were accidentally collected on the full set of images (training + testing sets), so accuracy is higher than it should be.

Rank Team Name Precision Recall F1-Score FPS Total Score
1 InvolutionNet 0.803 0.709 0.753 4.93 2.795354
2 FSBIN 0.015 0.003 0.005 2.3 0.000058