Releases: prushh/image-retrieval-fire
Releases · prushh/image-retrieval-fire
Project submission
Experiments are explained in the specific section of the README.md file.
The following archives refer to both experiments:
- freeze_backbone
- nnum_3
For each of them, you will find two archives related to the evaluation and one related to the training. The latter contains essential files such as saved models, the log, run configuration and plots.
The results obtained are reported for convenience in the following table, the scoring metric is the mAP:
Experiment | val_eccv20 | roxford5k | rparis6k | ||||
---|---|---|---|---|---|---|---|
Easy | Medium | Hard | Easy | Medium | Hard | ||
original | 90.08 | 98.1 | 82.77 | 62.6 | 95.45 | 85.11 | 69.87 |
freeze-backbone-best | 84.58 | 90.02 | 71.68 | 46.24 | 95 | 83.42 | 66.57 |
freeze-backbone-epoch200 | 71.91 | 67.95 | 50.92 | 24.16 | 88.67 | 72.79 | 49.51 |
nnum-3-best | 88.95 | 95.93 | 80.39 | 58.83 | 94.76 | 83.74 | 66.91 |
nnum-3-epoch200 | 88.52 | 96.6 | 79.97 | 57.61 | 94.64 | 83.87 | 67.87 |