ICMR2021 Grand Challenge: The PAIR competition ended successfully
The “ICMR2021 Grand Challenge – Embedded Deep Learning Object Detection Model Compression Competition for Traffic in Asian Countries” co-sponsored by MediaTek Inc., PAIR Labs and iVS Lab (Directed by Prof. Jiun-In Guo) has successfully concluded. This competition is mainly aimed at the unique transportation and road conditions in Asia. After two months of qualification competition and four weeks of final competition, the winners and special awards were finally released.
Award Winners
- Champion: as798792 (國立台南大學)
- First Runner-up: Deep Learner (北京交通大學)
- Second Runner-up: UCBH (國立成功大學電通所)
Special Awards
- Best accuracy award: as798792 (國立台南大學)
- Best bicycle detection award: as798792 (國立台南大學)
- Best scooter detection award: abcda (國立臺灣師範大學資工系)
Final Evaluation Result
| Team | Accuracy (%) | Model Size (MByte) | Complexity
(FLOPs) |
Speed
(s/image) |
||
| mAP | scooter | bicycle | ||||
| as798792 | 59 | 53.5 | 53.5 | 12 | 5.04G | 24.57 |
| Deep Learner | 53.2 | 51.2 | 31.5. | 28.53 | 2.46G | 18.82 |
| UCBH | 53.7 | 62 | 7.1 | 28.62 | 23.63G | 105.83 |
| yifu | 42.5 | 54.2 | 8.7 | 24.59 | 12.16G | 54.95 |
| pui pui | 52.4 | 63.1 | 0.3 | 84.72 | 64.32G | 60.18 |
| abcda | 56.8 | 63.3 | 14.6 | 192.36 | 390.83G | – |
| IRIS | 49.9 | 60.1 | 4.4 | 76.06 | – | 418.47 |
| lsm98 | 37.6 | 45.5 | – | 155 | – | 2497.1 |
| johnson0213 | – | – | – | 121 | 116.12G | – |
Final Score
| Team | Partial Score | Final Score | |||
| mAP | Size | Complexity | Speed | ||
| as798792 | 40 | 15 | 13 | 27 | 95 |
| Deep Learner | 25 | 11 | 15 | 30 | 81 |
| UCBH | 30 | 9 | 9 | 13 | 61 |
| yifu | 10 | 13 | 11 | 22 | 56 |
| pui pui | 20 | 6 | 6 | 18 | 50 |
| abcda | 35 | 0 | 2 | 0 | 37 |
| IRIS | 15 | 8 | 0 | 9 | 32 |
| lsm98 | 5 | 2 | 0 | 4 | 11 |
| johnson0213 | 0 | 4 | 4 | 0 | 8 |
