Challenge Winners
- Teams need to submit workshop paper to be eligible for awards
- Teams need to open-source full code for result reproduction to be eligible for awards
Track 1:
Winner: Team28 matcher
Box-Grained Reranking Matching for Multi-Camera Multi-Target Tracking
Runner-up: Team59 BOE
Multi-camera vehicle tracking system for AI City Challenge 2022
The following table shows the top teams (that have submitted papers) from the public leader board of Track 1 by the challenge submission deadline.
| Rank | Team ID | Team Name | Score |
| 1 | 28 | matcher | 0.8486 |
| 2 | 59 | BOE | 0.8437 |
| 3 | 37 | TAG | 0.8371 |
| 4 | 50 | FraunhoderIOSB | 0.8348 |
| 10 | 94 | SKKU | 0.8129 |
| 18 | 4 | HCMIU | 0.7255 |
Track 2:
Joint Winner: Team183 MegVideo
Symmetric Network with Spatial Relationship Modeling for Natural Language-based Vehicle Retrieval
Joint Winner: Team176 Must Win (Baidu-SYSU)
A multi-granularity retrieval system for natural language-based vehicle retrieval
Runner-up: Team91 HCMUS
Text Query based Traffic Video Event Retrieval with Global-Local Fusion Embedding
The following table shows the top teams from the public leader board of Track 2 by the challenge submission deadline.
- Team4 did not open-source their code for result reproduction thus per challenge rule we have to disqualify team4 for winning the award.
- Team176 used track1 data for training (which is prohibited) that gives a significant advantage by workshop time. Afterwards team176 removed track1 data from training and demonstrate their performance is still far ahead. Considering all above we grand team 176 and team 183 joined winners.
| Rank | Team ID | Team | Score (MRR) |
| 1 | 176 | Must Win | 0.6606 |
| 3 | 4 | HCMIU-CVIP | 0.4773 |
| 4 | 183 | MegVideo | 0.4392 |
| 5 | 91 | HCMUS | 0.3611 |
| 7 | 10 | Terminus-AI | 0.3320 |
| 9 | 24 | BUPT_MCPRL_T2 | 0.3012 |
Track 3:
Winner: Team72 VTCC-UTVM
An effective temporal localization method with multi-view 3D action recognition for untrimmed naturalistic driving videos
Runner-up: Team95 Tahakom
Temporal driver action recognition using action classification method
The following chart shows the performance of submitted codes from top teams on test set B:
| Rank | Team ID | Team | F-1 Score |
| 1 | 72 | VTCC-UTVM | 0.4025 |
| 2 | 95 | Tahakom | 0.3261 |
| 3 | 43 | Stargazer | 0.3152 |
| 4 | 1 | SCU_Anastasiu | 0.2381 |
| 5 | 16 | BUPT-MCPRL2 | 0.2143 |
Track 4:
Winner: Team9 CyberCore
Improving Domain Generalization by Learning without Forgetting: Application in Retail Checkout
Runner-up: Team117 GRAPH@FIT
Image inpainting for automated checkout solution
- Team55 did not submit paper to reveal their work thus per challenge rule we have to disqualify team55 for winning the award.
The following chart shows the performance of submitted codes from top teams on Dataset B. All submitted codes were tested on the same testing machine with the following specs:
