Benchmark for Image Retrieval (BKIR)
This project tries to build a benchmark for image retrieval, particully for Instance-level image retrieval.
Methods
The following methods are evaluated on Oxford Building dataset. The evaluation adopts mean Average Precision (mAP), which is computed using the code provided by compute_ap.cpp.
| method | feature | mAP (best) | status | evalute code |
|---|---|---|---|---|
| fc_retrieval | CNN | 60.2% | finished | fc_retrieval |
| rmac_retrieval | RMAC | 75.7%(256d, crop, qe) | finished | rmac_retrieval |
| crow_retrieval | CROW | 72.8%(256d, crop, qe) | finished | crow_retrieval |
| fv_retrieval | SIFT | 67.29% | finished | fv_retrieval |
| vlad_retrieval | SIFT | 63.13% | finished | vlad_retrieval |
| fv_retrieval | SOSNet | 50.73% | ongoing | - |
| vlad_retrieval | SOSNet | - | ongoing | - |
the methods on above have the following characteristics:
- Low dimension
- Time - tested, and are dimanstracted effectively
- Used in industry
Contribution
If you are interested in this project, feel free to contribute your code. Only Python and C++ code are accepted.