Command Palette
Search for a command to run...
Alexander Hermans; Lucas Beyer; Bastian Leibe
Abstract
In the past few years, the field of computer vision has gone through a revolution fueled mainly by the advent of large datasets and the adoption of deep convolutional neural networks for end-to-end learning. The person re-identification subfield is no exception to this. Unfortunately, a prevailing belief in the community seems to be that the triplet loss is inferior to using surrogate losses (classification, verification) followed by a separate metric learning step. We show that, for models trained from scratch as well as pretrained ones, using a variant of the triplet loss to perform end-to-end deep metric learning outperforms most other published methods by a large margin.
Code Repositories
Benchmarks
| Benchmark | Methodology | Metrics |
|---|---|---|
| person-re-identification-on-cuhk03 | TriNet | Rank-1: 89.63 Rank-5: 99.01 |
| person-re-identification-on-dukemtmc-reid | TriNet | Rank-1: 72.44 mAP: 53.50 |
| person-re-identification-on-market-1501 | LuNet | Rank-1: 81.38 Rank-5: 92.34 mAP: 60.71 |
| person-re-identification-on-market-1501 | TriNet (RK) | Rank-1: 86.67 Rank-5: 93.38 mAP: 81.07 |
| person-re-identification-on-market-1501 | LuNet (RK) | Rank-1: 84.59 Rank-5: 91.89 mAP: 75.62 |
| person-re-identification-on-market-1501 | TriNet | Rank-1: 84.92 Rank-5: 94.21 mAP: 69.14 |
| person-re-identification-on-mars | TriNet | Rank-1: 79.80 Rank-5: 91.36 mAP: 67.70 |
| person-re-identification-on-mars | LuNet (RK) | Rank-1: 78.48 Rank-5: 88.74 mAP: 73.68 |
| person-re-identification-on-mars | TriNet (RK) | Rank-1: 81.21 Rank-5: 90.76 mAP: 77.43 |
| person-re-identification-on-mars | LuNet | Rank-1: 75.56 Rank-5: 89.70 mAP: 60.48 |
Build AI with AI
From idea to launch — accelerate your AI development with free AI co-coding, out-of-the-box environment and best price of GPUs.