Adhiraj Ghosh, Kuruparan Shanmugalingam, Wen-Yan Lin
Object appearances change dramatically with pose variations. This creates a challenge for embedding schemes that seek to map instances with the same object ID to locations that are as close as possible. This issue becomes significantly heightened in complex computer vision tasks such as re-identification(reID). In this paper, we suggest that these dramatic appearance changes are indications that an object ID is composed of multiple natural groups, and it is counterproductive to forcefully map instances from different groups to a common location. This leads us to introduce Relation Preserving Triplet Mining (RPTM), a feature-matching guided triplet mining scheme, that ensures that triplets will respect the natural subgroupings within an object ID. We use this triplet mining mechanism to establish a pose-aware, well-conditioned triplet loss by implicitly enforcing view consistency. This allows a single network to be trained with fixed parameters across datasets while providing state-of-the-art results. Code is available at https://github.com/adhirajghosh/RPTM_reid.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Person Re-Identification | DukeMTMC-reID | Rank-1 | 93.5 | RPTM |
| Person Re-Identification | DukeMTMC-reID | Rank-5 | 96.1 | RPTM |
| Person Re-Identification | DukeMTMC-reID | mAP | 89.2 | RPTM |
| Intelligent Surveillance | VehicleID Large | Rank-1 | 92.9 | RPTM |
| Intelligent Surveillance | VehicleID Large | Rank-5 | 96.3 | RPTM |
| Intelligent Surveillance | VehicleID Large | mAP | 80.5 | RPTM |
| Intelligent Surveillance | VehicleID Medium | Rank-1 | 93.3 | RPTM |
| Intelligent Surveillance | VehicleID Medium | Rank-5 | 96.5 | RPTM |
| Intelligent Surveillance | VehicleID Medium | mAP | 81.2 | RPTM |
| Intelligent Surveillance | VeRi-776 | Rank-1 | 97.3 | RPTM |
| Intelligent Surveillance | VeRi-776 | Rank1 | 97.3 | RPTM |
| Intelligent Surveillance | VeRi-776 | Rank5 | 98.4 | RPTM |
| Intelligent Surveillance | VeRi-776 | mAP | 88 | RPTM |
| Intelligent Surveillance | VehicleID Small | Rank-1 | 95.5 | RPTM |
| Intelligent Surveillance | VehicleID Small | Rank-5 | 97.4 | RPTM |
| Intelligent Surveillance | VehicleID Small | mAP | 84.8 | RPTM |
| Vehicle Re-Identification | VehicleID Large | Rank-1 | 92.9 | RPTM |
| Vehicle Re-Identification | VehicleID Large | Rank-5 | 96.3 | RPTM |
| Vehicle Re-Identification | VehicleID Large | mAP | 80.5 | RPTM |
| Vehicle Re-Identification | VehicleID Medium | Rank-1 | 93.3 | RPTM |
| Vehicle Re-Identification | VehicleID Medium | Rank-5 | 96.5 | RPTM |
| Vehicle Re-Identification | VehicleID Medium | mAP | 81.2 | RPTM |
| Vehicle Re-Identification | VeRi-776 | Rank-1 | 97.3 | RPTM |
| Vehicle Re-Identification | VeRi-776 | Rank1 | 97.3 | RPTM |
| Vehicle Re-Identification | VeRi-776 | Rank5 | 98.4 | RPTM |
| Vehicle Re-Identification | VeRi-776 | mAP | 88 | RPTM |
| Vehicle Re-Identification | VehicleID Small | Rank-1 | 95.5 | RPTM |
| Vehicle Re-Identification | VehicleID Small | Rank-5 | 97.4 | RPTM |
| Vehicle Re-Identification | VehicleID Small | mAP | 84.8 | RPTM |