| Methods | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | Apmk | AP | AP | Apbb | AP AP | Apmk | AP | AP | |||
| Scratch | - | 31.0 | 49.5 | 33.2 | 28.5 | 46.8 | 30.4 | 38.4 | 57.5 | 42.0 | 34.7 | 54.8 | 37.2 |
| Supervised | 90 | 38.9 | 59.6 | 42.7 | 35.4 | 56.5 | 38.1 | 41.3 | 61.3 | 45.0 | 37.3 | 58.3 | 40.3 |
| MoCo[4] | 200 | 38.5 | 58.9 | 42.0 | 35.1 | 55.9 | 37.7 | 40.8 | 61.6 | 44.7 | 36.9 | 58.4 | 39.7 |
| MoCo v2[5] | 200 | 40.4 | 60.2 | 44.2 | 36.4 | 57.2 | 38.9 | 41.7 | 61.6 | 45.6 | 37.6 | 58.7 | 40.5 |
| InfoMin [6] | 200 | 40.6 | 60.6 | 44.6 | 36.7 | 57.7 | 39.4 | 42.5 | 62.7 | 46.8 | 38.4 | 59.7 | 41.4 |
| BYOL[3] | 300 | 40.4 | 61.6 | 44.1 | 37.2 | 58.8 | 39.8 | 42.3 | 62.6 | 46.2 | 38.3 | 59.6 | 41.1 |
| SwAV[7] | 400 | - | - | - | - | - | - | 42.3 | 62.8 | 46.3 | 38.2 | 60.0 | 41.0 |
| ReSim-FPNT [45] | 200 | 39.8 | 60.2 | 43.5 | 36.0 | 57.1 | 38.6 | 41.4 | 61.9 | 45.4 | 37.5 | 59.1 | 40.3 |
| PixPro[10] | 400 | 41.4 | 61.6 | 45.4 | 1 | - | 1 | - | - | - | - | 1 | - |
| InsLoc [12] | 400 | 42.0 | 62.3 | 45.8 | 37.6 | 59.0 | 40.5 | 43.3 | 63.6 | 47.3 | 38.8 | 60.9 | 41.7 |
| DenseCL[11] | 200 | 40.3 | 59.9 | 44.3 | 36.4 | 57.0 | 39.2 | 41.2 | 61.9 | 45.1 | 37.3 | 58.9 | 40.1 |
| DetCons [13] | 1000 | 41.8 | 1 | 37.4 | - | ： | 42.9 | 1 | - | 38.1 | ： | ||
| DetConB [13] | 1000 | 42.7 | - | 38.2 | 1 | ： | 43.4 | 1 | - | 38.7 | - | ： | |
| SoCo | 100 | 42.3 | 62.5 | 46.5 | 37.6 | 59.1 | 40.5 | 43.2 | 63.3 | 47.3 | 38.8 | 60.6 | 41.9 |
| SoCo | 400 | 43.0 | 63.3 | 47.1 | 38.2 | 60.2 | 41.0 | 44.0 | 64.0 | 48.4 | 39.0 | 61.3 | 41.7 |
| SoCo* | 400 | 43.2 | 63.5 | 47.4 | 38.4 | 60.2 | 41.4 | 44.3 | 64.6 | 48.9 | 39.6 | 61.8 | 42.5 |
| Methods | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | Apmk | AP | APP | Apbb | AP8 | AP | Apmk | AP | APP | ||
| Scratch | - | 26.4 | 44.0 | 27.8 | 29.3 | 46.9 | 30.8 | 35.6 | 54.6 | 38.2 | 31.4 | 51.5 | 33.5 |
| Supervised | 90 | 38.2 | 58.2 | 41.2 | 33.3 | 54.7 | 35.2 | 40.0 | 59.9 | 43.1 | 34.7 | 56.5 | 36.9 |
| MoCo [4] | 200 | 38.5 | 58.3 | 41.6 | 33.6 | 54.8 | 35.6 | 40.7 | 60.5 | 44.1 | 35.4 | 57.3 | 37.6 37.1 |
| SimCLR [9] MoCo v2 [5] | 200 800 | - 39.3 | - | ■ | - | - | 1 | 39.6 41.2 | 59.1 60.9 | 42.9 44.6 | 34.6 35.8 | 55.9 57.7 | 38.2 |
| InfoMin 6] | 58.9 | 42.5 | 34.3 | 55.7 | 36.5 | 61.2 | 45.0 | 57.9 | 38.3 | ||||
| BYOL[3] | 200 300 | 39.0 | 58.5 | 42.0 | 34.1 | 55.2 | 36.3 | 41.3 | 36.0 | 56.8 | 37.3 | ||
| SwAV[7] | 400 | - | - | - | - | - | - | 40.3 | 60.5 | 43.9 | 35.1 | 36.6 | |
| SimSiam [8] | 200 | - 39.2 | - 59.3 | - | 1 | 1 | - | 39.6 | 60.1 | 42.9 | 34.7 | 56.6 | |
| PixPro[10] | 400 | 40.5 | 59.8 | 42.1 | 34.4 | 56.0 | 36.7 | - | ： | - | - | - | - |
| InsLoc [12] | 400 | 39.8 | 59.6 | 44.0 42.9 | - | - | - | - 41.8 | - 61.6 | - | - | - 58.2 | - 38.8 |
| 34.7 | 56.3 | 36.9 | 45.4 | 36.3 | |||||||||
| SoCo | 100 | 40.4 | 60.4 | 43.7 | 34.9 | 56.8 | 37.0 | 41.1 | 61.0 | 44.4 | 35.6 | 57.5 | 38.0 |
| SoCo | 400 | 40.9 | 60.9 | 44.3 | 35.3 | 57.5 | 37.3 | 42.0 | 61.8 | 45.6 | 36.3 | 58.5 | 38.8 |
| (a) Sub-table 1. | ||||
| Methods | Epoch | APbb | AP | AP |
| Scratch Supervised | - 90 | 33.8 53.5 | 60.2 81.3 | 33.1 58.8 |
| ReSim-C4 [45] PixPro [10] InsLoc [12] DenseCL[11] | 200 400 400 200 | 58.7 60.2 58.4 58.7 | 83.1 83.8 83.0 82.8 | 66.3 67.7 65.3 65.2 |
| SoCo SoCo | 100 400 | 59.1 59.7 | 83.4 83.8 | 65.6 66.8 |
| Methods | Epoch | Apbb | AP | AP |
| MoCo[4] | 200 | 55.9 | 81.5 | 62.6 |
| SimCLR [9] | 1000 | 56.3 | 81.9 | 62.5 |
| MoCo v2[5] | 800 | 57.6 | 82.7 | 64.4 |
| InfoMin [6] | 200 | 57.6 | 82.7 | 64.6 |
| BYOL[3] | 300 | 51.9 | 81.0 | 56.5 |
| SwAV[7] | 400 | 45.1 | 77.4 | 46.5 |
| SimSiam[8] | 200 | 57.0 | 82.4 | 63.7 |
| ReSim-FPN[45] | 200 | 59.2 | 82.9 | 65.9 |
| Whole Image | Selective Search | FPN | Head | Scale-aware Assignment | Box Jitter | Multi View | Apbb | Apmk |
| √ | 38.1 | 34.4 | ||||||
| √ | 40.6 (+2.5) | 36.8 (+2.4) | ||||||
| √ | 40.2 (+2.1) | 36.2 (+1.8) | ||||||
| √ | 广 | 41.2 (+3.1) | 37.0 (+2.6) | |||||
| √ | 1 | 41.6 (+3.5) | 37.3 (+2.9) | |||||
| √ | >>>>>> | >>>>> | 1 | 4 | 41.7 (+3.6) | 37.5 (+3.1) | ||
| √ | 厂 | 厂 | 42.3 (+4.2) | 37.6 (+3.2) |
| Image Size | Apbb | Apmk |
| 96 | 42.1 | 37.7 |
| 112 | 42.3 | 37.6 |
| 128 | 42.1 | 37.7 |
| 160 | 42.0 | 37.6 |
| 192 | 42.2 | 37.8 |
| Batch Size | APbb | Apmk |
| 512 | 41.7 | 37.6 |
| 1024 | 41.9 | 37.6 |
| 2048 | 42.3 | 37.6 |
| 4096 | 41.4 | 37.3 |
| Selective Search | Random | K | APbb | Apmk |
| < | 1 | 41.6 | 37.3 | |
| 486 | 42.3 41.6 | 37.6 | ||
| √ | 37.4 | |||
| √ | 41.2 | 37.0 | ||
| >>> | 148 | 41.4 | 36.9 | |
| NaN NaN | NaN NaN |
| T | APbb | Apmk |
| 0.98 | 35.0 | 31.7 |
| 0.99 | 42.3 | 37.6 |
| 0.993 | 41.8 | 37.6 |
| Method | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | APmk | AP | AP7 | Apbb AP | AP | Apmk | AP | AP | |||
| Supervised | 90 | 20.4 | 32.9 | 21.7 | 19.4 | 30.6 | 20.5 | 23.4 | 36.9 | 24.9 | 22.3 | 34.7 | 23.5 |
| SoCo* | 400 | 26.3 | 41.2 | 27.8 | 25.0 | 38.5 | 26.8 | 28.3 | 43.5 | 30.7 | 26.9 | 41.1 | 28.7 |
| Method | Epoch | Apbb | AP8 | AP |
| Supervised | 90 | 36.3 | 55.3 | 38.6 |
| SoCo* 一 | 400 | 38.3 | 57.2 | 41.2 |
| Method | Epoch | Apbb | AP8 | AP9 |
| Supervised | 90 | 36.6 | 56.0 | 38.8 |
| SoCo* | 400 | 37.4 | 56.3 | 39.9 |
| Methods | Epoch | Mini COCO (5%) | Mini C0CO (10%) | ||||||||||
| Apbb AP8 | AP9 | APmk | APk | APm | Apbb | AP8 | AP | Apmk | AP APP | ||||
| Supervised | 90 | 19.4 | 36.6 | 18.6 | 18.3 | 33.5 | 17.9 | 24.7 | 43.1 | 25.3 | 22.9 | 40.0 | 23.4 |
| SoCo | 100 | 24.6 | 42.2 | 25.6 | 22.1 | 38.8 | 22.4 | 29.2 | 47.7 | 31.1 | 26.1 | 44.4 | 27.0 |
| SoCo | 400 | 26.0 | 43.2 | 27.5 | 22.9 | 40.0 | 23.4 | 30.4 | 48.6 | 32.5 | 26.8 | 45.2 | 27.9 |
| SoCo* | 400 | 26.8 | 45.0 | 28.3 | 23.8 | 41.4 | 24.2 | 31.1 | 49.9 | 33.4 | 27.6 | 46.4 | 28.9 |
| Pretraining dataset | Epoch | Apbb | AP | AP | Apmk | AP | APP |
| ImageNet | 100 | 42.3 | 62.5 | 46.5 | 37.6 | 59.1 | 40.5 |
| ImageNet-subset | 100 | 36.9 | 56.2 | 40.1 | 33.2 | 53.0 | 35.6 |
| COCO train set + unlabeled set | 100 | 37.3 | 56.5 | 40.6 | 33.5 | 53.4 | 36.0 |
| COCO train set + unlabeled set | 530 | 40.6 | 61.1 | 44.4 | 36.4 | 58.1 | 38.7 |
| Method | Epoch | APbb | APb 50 | AP | Apmk | AP 50 | APP |
| Supervised | 90 | 41.9 | 61.5 | 45.4 | 37.7 | 58.8 | 40.5 |
| SoCo* | 400 | 44.5 | 64.2 | 48.8 | 39.6 | 61.5 | 42.4 |
| Methods | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | Apmk | AP | AP | Apbb | AP AP | Apmk | AP | AP | |||
| Scratch | - | 31.0 | 49.5 | 33.2 | 28.5 | 46.8 | 30.4 | 38.4 | 57.5 | 42.0 | 34.7 | 54.8 | 37.2 |
| Supervised | 90 | 38.9 | 59.6 | 42.7 | 35.4 | 56.5 | 38.1 | 41.3 | 61.3 | 45.0 | 37.3 | 58.3 | 40.3 |
| MoCo[4] | 200 | 38.5 | 58.9 | 42.0 | 35.1 | 55.9 | 37.7 | 40.8 | 61.6 | 44.7 | 36.9 | 58.4 | 39.7 |
| MoCo v2[5] | 200 | 40.4 | 60.2 | 44.2 | 36.4 | 57.2 | 38.9 | 41.7 | 61.6 | 45.6 | 37.6 | 58.7 | 40.5 |
| InfoMin [6] | 200 | 40.6 | 60.6 | 44.6 | 36.7 | 57.7 | 39.4 | 42.5 | 62.7 | 46.8 | 38.4 | 59.7 | 41.4 |
| BYOL[3] | 300 | 40.4 | 61.6 | 44.1 | 37.2 | 58.8 | 39.8 | 42.3 | 62.6 | 46.2 | 38.3 | 59.6 | 41.1 |
| SwAV[7] | 400 | - | - | - | - | - | - | 42.3 | 62.8 | 46.3 | 38.2 | 60.0 | 41.0 |
| ReSim-FPNT [45] | 200 | 39.8 | 60.2 | 43.5 | 36.0 | 57.1 | 38.6 | 41.4 | 61.9 | 45.4 | 37.5 | 59.1 | 40.3 |
| PixPro[10] | 400 | 41.4 | 61.6 | 45.4 | 1 | - | 1 | - | - | - | - | 1 | - |
| InsLoc [12] | 400 | 42.0 | 62.3 | 45.8 | 37.6 | 59.0 | 40.5 | 43.3 | 63.6 | 47.3 | 38.8 | 60.9 | 41.7 |
| DenseCL[11] | 200 | 40.3 | 59.9 | 44.3 | 36.4 | 57.0 | 39.2 | 41.2 | 61.9 | 45.1 | 37.3 | 58.9 | 40.1 |
| DetCons [13] | 1000 | 41.8 | 1 | 37.4 | - | ： | 42.9 | 1 | - | 38.1 | ： | ||
| DetConB [13] | 1000 | 42.7 | - | 38.2 | 1 | ： | 43.4 | 1 | - | 38.7 | - | ： | |
| SoCo | 100 | 42.3 | 62.5 | 46.5 | 37.6 | 59.1 | 40.5 | 43.2 | 63.3 | 47.3 | 38.8 | 60.6 | 41.9 |
| SoCo | 400 | 43.0 | 63.3 | 47.1 | 38.2 | 60.2 | 41.0 | 44.0 | 64.0 | 48.4 | 39.0 | 61.3 | 41.7 |
| SoCo* | 400 | 43.2 | 63.5 | 47.4 | 38.4 | 60.2 | 41.4 | 44.3 | 64.6 | 48.9 | 39.6 | 61.8 | 42.5 |
| Methods | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | Apmk | AP | APP | Apbb | AP8 | AP | Apmk | AP | APP | ||
| Scratch | - | 26.4 | 44.0 | 27.8 | 29.3 | 46.9 | 30.8 | 35.6 | 54.6 | 38.2 | 31.4 | 51.5 | 33.5 |
| Supervised | 90 | 38.2 | 58.2 | 41.2 | 33.3 | 54.7 | 35.2 | 40.0 | 59.9 | 43.1 | 34.7 | 56.5 | 36.9 |
| MoCo [4] | 200 | 38.5 | 58.3 | 41.6 | 33.6 | 54.8 | 35.6 | 40.7 | 60.5 | 44.1 | 35.4 | 57.3 | 37.6 37.1 |
| SimCLR [9] MoCo v2 [5] | 200 800 | - 39.3 | - | ■ | - | - | 1 | 39.6 41.2 | 59.1 60.9 | 42.9 44.6 | 34.6 35.8 | 55.9 57.7 | 38.2 |
| InfoMin 6] | 58.9 | 42.5 | 34.3 | 55.7 | 36.5 | 61.2 | 45.0 | 57.9 | 38.3 | ||||
| BYOL[3] | 200 300 | 39.0 | 58.5 | 42.0 | 34.1 | 55.2 | 36.3 | 41.3 | 36.0 | 56.8 | 37.3 | ||
| SwAV[7] | 400 | - | - | - | - | - | - | 40.3 | 60.5 | 43.9 | 35.1 | 36.6 | |
| SimSiam [8] | 200 | - 39.2 | - 59.3 | - | 1 | 1 | - | 39.6 | 60.1 | 42.9 | 34.7 | 56.6 | |
| PixPro[10] | 400 | 40.5 | 59.8 | 42.1 | 34.4 | 56.0 | 36.7 | - | ： | - | - | - | - |
| InsLoc [12] | 400 | 39.8 | 59.6 | 44.0 42.9 | - | - | - | - 41.8 | - 61.6 | - | - | - 58.2 | - 38.8 |
| 34.7 | 56.3 | 36.9 | 45.4 | 36.3 | |||||||||
| SoCo | 100 | 40.4 | 60.4 | 43.7 | 34.9 | 56.8 | 37.0 | 41.1 | 61.0 | 44.4 | 35.6 | 57.5 | 38.0 |
| SoCo | 400 | 40.9 | 60.9 | 44.3 | 35.3 | 57.5 | 37.3 | 42.0 | 61.8 | 45.6 | 36.3 | 58.5 | 38.8 |
| (a) Sub-table 1. | ||||
| Methods | Epoch | APbb | AP | AP |
| Scratch Supervised | - 90 | 33.8 53.5 | 60.2 81.3 | 33.1 58.8 |
| ReSim-C4 [45] PixPro [10] InsLoc [12] DenseCL[11] | 200 400 400 200 | 58.7 60.2 58.4 58.7 | 83.1 83.8 83.0 82.8 | 66.3 67.7 65.3 65.2 |
| SoCo SoCo | 100 400 | 59.1 59.7 | 83.4 83.8 | 65.6 66.8 |
| Methods | Epoch | Apbb | AP | AP |
| MoCo[4] | 200 | 55.9 | 81.5 | 62.6 |
| SimCLR [9] | 1000 | 56.3 | 81.9 | 62.5 |
| MoCo v2[5] | 800 | 57.6 | 82.7 | 64.4 |
| InfoMin [6] | 200 | 57.6 | 82.7 | 64.6 |
| BYOL[3] | 300 | 51.9 | 81.0 | 56.5 |
| SwAV[7] | 400 | 45.1 | 77.4 | 46.5 |
| SimSiam[8] | 200 | 57.0 | 82.4 | 63.7 |
| ReSim-FPN[45] | 200 | 59.2 | 82.9 | 65.9 |
| Whole Image | Selective Search | FPN | Head | Scale-aware Assignment | Box Jitter | Multi View | Apbb | Apmk |
| √ | 38.1 | 34.4 | ||||||
| √ | 40.6 (+2.5) | 36.8 (+2.4) | ||||||
| √ | 40.2 (+2.1) | 36.2 (+1.8) | ||||||
| √ | 广 | 41.2 (+3.1) | 37.0 (+2.6) | |||||
| √ | 1 | 41.6 (+3.5) | 37.3 (+2.9) | |||||
| √ | >>>>>> | >>>>> | 1 | 4 | 41.7 (+3.6) | 37.5 (+3.1) | ||
| √ | 厂 | 厂 | 42.3 (+4.2) | 37.6 (+3.2) |
| Image Size | Apbb | Apmk |
| 96 | 42.1 | 37.7 |
| 112 | 42.3 | 37.6 |
| 128 | 42.1 | 37.7 |
| 160 | 42.0 | 37.6 |
| 192 | 42.2 | 37.8 |
| Batch Size | APbb | Apmk |
| 512 | 41.7 | 37.6 |
| 1024 | 41.9 | 37.6 |
| 2048 | 42.3 | 37.6 |
| 4096 | 41.4 | 37.3 |
| Selective Search | Random | K | APbb | Apmk |
| < | 1 | 41.6 | 37.3 | |
| 486 | 42.3 41.6 | 37.6 | ||
| √ | 37.4 | |||
| √ | 41.2 | 37.0 | ||
| >>> | 148 | 41.4 | 36.9 | |
| NaN NaN | NaN NaN |
| T | APbb | Apmk |
| 0.98 | 35.0 | 31.7 |
| 0.99 | 42.3 | 37.6 |
| 0.993 | 41.8 | 37.6 |
| Method | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | APmk | AP | AP7 | Apbb AP | AP | Apmk | AP | AP | |||
| Supervised | 90 | 20.4 | 32.9 | 21.7 | 19.4 | 30.6 | 20.5 | 23.4 | 36.9 | 24.9 | 22.3 | 34.7 | 23.5 |
| SoCo* | 400 | 26.3 | 41.2 | 27.8 | 25.0 | 38.5 | 26.8 | 28.3 | 43.5 | 30.7 | 26.9 | 41.1 | 28.7 |
| Method | Epoch | Apbb | AP8 | AP |
| Supervised | 90 | 36.3 | 55.3 | 38.6 |
| SoCo* 一 | 400 | 38.3 | 57.2 | 41.2 |
| Method | Epoch | Apbb | AP8 | AP9 |
| Supervised | 90 | 36.6 | 56.0 | 38.8 |
| SoCo* | 400 | 37.4 | 56.3 | 39.9 |
| Methods | Epoch | Mini COCO (5%) | Mini C0CO (10%) | ||||||||||
| Apbb AP8 | AP9 | APmk | APk | APm | Apbb | AP8 | AP | Apmk | AP APP | ||||
| Supervised | 90 | 19.4 | 36.6 | 18.6 | 18.3 | 33.5 | 17.9 | 24.7 | 43.1 | 25.3 | 22.9 | 40.0 | 23.4 |
| SoCo | 100 | 24.6 | 42.2 | 25.6 | 22.1 | 38.8 | 22.4 | 29.2 | 47.7 | 31.1 | 26.1 | 44.4 | 27.0 |
| SoCo | 400 | 26.0 | 43.2 | 27.5 | 22.9 | 40.0 | 23.4 | 30.4 | 48.6 | 32.5 | 26.8 | 45.2 | 27.9 |
| SoCo* | 400 | 26.8 | 45.0 | 28.3 | 23.8 | 41.4 | 24.2 | 31.1 | 49.9 | 33.4 | 27.6 | 46.4 | 28.9 |
| Pretraining dataset | Epoch | Apbb | AP | AP | Apmk | AP | APP |
| ImageNet | 100 | 42.3 | 62.5 | 46.5 | 37.6 | 59.1 | 40.5 |
| ImageNet-subset | 100 | 36.9 | 56.2 | 40.1 | 33.2 | 53.0 | 35.6 |
| COCO train set + unlabeled set | 100 | 37.3 | 56.5 | 40.6 | 33.5 | 53.4 | 36.0 |
| COCO train set + unlabeled set | 530 | 40.6 | 61.1 | 44.4 | 36.4 | 58.1 | 38.7 |
| Method | Epoch | APbb | APb 50 | AP | Apmk | AP 50 | APP |
| Supervised | 90 | 41.9 | 61.5 | 45.4 | 37.7 | 58.8 | 40.5 |
| SoCo* | 400 | 44.5 | 64.2 | 48.8 | 39.6 | 61.5 | 42.4 |
| Methods | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | Apmk | AP | AP | Apbb | AP AP | Apmk | AP | AP | |||
| Scratch | - | 31.0 | 49.5 | 33.2 | 28.5 | 46.8 | 30.4 | 38.4 | 57.5 | 42.0 | 34.7 | 54.8 | 37.2 |
| Supervised | 90 | 38.9 | 59.6 | 42.7 | 35.4 | 56.5 | 38.1 | 41.3 | 61.3 | 45.0 | 37.3 | 58.3 | 40.3 |
| MoCo[4] | 200 | 38.5 | 58.9 | 42.0 | 35.1 | 55.9 | 37.7 | 40.8 | 61.6 | 44.7 | 36.9 | 58.4 | 39.7 |
| MoCo v2[5] | 200 | 40.4 | 60.2 | 44.2 | 36.4 | 57.2 | 38.9 | 41.7 | 61.6 | 45.6 | 37.6 | 58.7 | 40.5 |
| InfoMin [6] | 200 | 40.6 | 60.6 | 44.6 | 36.7 | 57.7 | 39.4 | 42.5 | 62.7 | 46.8 | 38.4 | 59.7 | 41.4 |
| BYOL[3] | 300 | 40.4 | 61.6 | 44.1 | 37.2 | 58.8 | 39.8 | 42.3 | 62.6 | 46.2 | 38.3 | 59.6 | 41.1 |
| SwAV[7] | 400 | - | - | - | - | - | - | 42.3 | 62.8 | 46.3 | 38.2 | 60.0 | 41.0 |
| ReSim-FPNT [45] | 200 | 39.8 | 60.2 | 43.5 | 36.0 | 57.1 | 38.6 | 41.4 | 61.9 | 45.4 | 37.5 | 59.1 | 40.3 |
| PixPro[10] | 400 | 41.4 | 61.6 | 45.4 | 1 | - | 1 | - | - | - | - | 1 | - |
| InsLoc [12] | 400 | 42.0 | 62.3 | 45.8 | 37.6 | 59.0 | 40.5 | 43.3 | 63.6 | 47.3 | 38.8 | 60.9 | 41.7 |
| DenseCL[11] | 200 | 40.3 | 59.9 | 44.3 | 36.4 | 57.0 | 39.2 | 41.2 | 61.9 | 45.1 | 37.3 | 58.9 | 40.1 |
| DetCons [13] | 1000 | 41.8 | 1 | 37.4 | - | ： | 42.9 | 1 | - | 38.1 | ： | ||
| DetConB [13] | 1000 | 42.7 | - | 38.2 | 1 | ： | 43.4 | 1 | - | 38.7 | - | ： | |
| SoCo | 100 | 42.3 | 62.5 | 46.5 | 37.6 | 59.1 | 40.5 | 43.2 | 63.3 | 47.3 | 38.8 | 60.6 | 41.9 |
| SoCo | 400 | 43.0 | 63.3 | 47.1 | 38.2 | 60.2 | 41.0 | 44.0 | 64.0 | 48.4 | 39.0 | 61.3 | 41.7 |
| SoCo* | 400 | 43.2 | 63.5 | 47.4 | 38.4 | 60.2 | 41.4 | 44.3 | 64.6 | 48.9 | 39.6 | 61.8 | 42.5 |
| Methods | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | Apmk | AP | AP | Apbb | AP AP | Apmk | AP | AP | |||
| Scratch | - | 31.0 | 49.5 | 33.2 | 28.5 | 46.8 | 30.4 | 38.4 | 57.5 | 42.0 | 34.7 | 54.8 | 37.2 |
| Supervised | 90 | 38.9 | 59.6 | 42.7 | 35.4 | 56.5 | 38.1 | 41.3 | 61.3 | 45.0 | 37.3 | 58.3 | 40.3 |
| MoCo[4] | 200 | 38.5 | 58.9 | 42.0 | 35.1 | 55.9 | 37.7 | 40.8 | 61.6 | 44.7 | 36.9 | 58.4 | 39.7 |
| MoCo v2[5] | 200 | 40.4 | 60.2 | 44.2 | 36.4 | 57.2 | 38.9 | 41.7 | 61.6 | 45.6 | 37.6 | 58.7 | 40.5 |
| InfoMin [6] | 200 | 40.6 | 60.6 | 44.6 | 36.7 | 57.7 | 39.4 | 42.5 | 62.7 | 46.8 | 38.4 | 59.7 | 41.4 |
| BYOL[3] | 300 | 40.4 | 61.6 | 44.1 | 37.2 | 58.8 | 39.8 | 42.3 | 62.6 | 46.2 | 38.3 | 59.6 | 41.1 |
| SwAV[7] | 400 | - | - | - | - | - | - | 42.3 | 62.8 | 46.3 | 38.2 | 60.0 | 41.0 |
| ReSim-FPNT [45] | 200 | 39.8 | 60.2 | 43.5 | 36.0 | 57.1 | 38.6 | 41.4 | 61.9 | 45.4 | 37.5 | 59.1 | 40.3 |
| PixPro[10] | 400 | 41.4 | 61.6 | 45.4 | 1 | - | 1 | - | - | - | - | 1 | - |
| InsLoc [12] | 400 | 42.0 | 62.3 | 45.8 | 37.6 | 59.0 | 40.5 | 43.3 | 63.6 | 47.3 | 38.8 | 60.9 | 41.7 |
| DenseCL[11] | 200 | 40.3 | 59.9 | 44.3 | 36.4 | 57.0 | 39.2 | 41.2 | 61.9 | 45.1 | 37.3 | 58.9 | 40.1 |
| DetCons [13] | 1000 | 41.8 | 1 | 37.4 | - | ： | 42.9 | 1 | - | 38.1 | ： | ||
| DetConB [13] | 1000 | 42.7 | - | 38.2 | 1 | ： | 43.4 | 1 | - | 38.7 | - | ： | |
| SoCo | 100 | 42.3 | 62.5 | 46.5 | 37.6 | 59.1 | 40.5 | 43.2 | 63.3 | 47.3 | 38.8 | 60.6 | 41.9 |
| SoCo | 400 | 43.0 | 63.3 | 47.1 | 38.2 | 60.2 | 41.0 | 44.0 | 64.0 | 48.4 | 39.0 | 61.3 | 41.7 |
| SoCo* | 400 | 43.2 | 63.5 | 47.4 | 38.4 | 60.2 | 41.4 | 44.3 | 64.6 | 48.9 | 39.6 | 61.8 | 42.5 |
| Methods | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | Apmk | AP | APP | Apbb | AP8 | AP | Apmk | AP | APP | ||
| Scratch | - | 26.4 | 44.0 | 27.8 | 29.3 | 46.9 | 30.8 | 35.6 | 54.6 | 38.2 | 31.4 | 51.5 | 33.5 |
| Supervised | 90 | 38.2 | 58.2 | 41.2 | 33.3 | 54.7 | 35.2 | 40.0 | 59.9 | 43.1 | 34.7 | 56.5 | 36.9 |
| MoCo [4] | 200 | 38.5 | 58.3 | 41.6 | 33.6 | 54.8 | 35.6 | 40.7 | 60.5 | 44.1 | 35.4 | 57.3 | 37.6 37.1 |
| SimCLR [9] MoCo v2 [5] | 200 800 | - 39.3 | - | ■ | - | - | 1 | 39.6 41.2 | 59.1 60.9 | 42.9 44.6 | 34.6 35.8 | 55.9 57.7 | 38.2 |
| InfoMin 6] | 58.9 | 42.5 | 34.3 | 55.7 | 36.5 | 61.2 | 45.0 | 57.9 | 38.3 | ||||
| BYOL[3] | 200 300 | 39.0 | 58.5 | 42.0 | 34.1 | 55.2 | 36.3 | 41.3 | 36.0 | 56.8 | 37.3 | ||
| SwAV[7] | 400 | - | - | - | - | - | - | 40.3 | 60.5 | 43.9 | 35.1 | 36.6 | |
| SimSiam [8] | 200 | - 39.2 | - 59.3 | - | 1 | 1 | - | 39.6 | 60.1 | 42.9 | 34.7 | 56.6 | |
| PixPro[10] | 400 | 40.5 | 59.8 | 42.1 | 34.4 | 56.0 | 36.7 | - | ： | - | - | - | - |
| InsLoc [12] | 400 | 39.8 | 59.6 | 44.0 42.9 | - | - | - | - 41.8 | - 61.6 | - | - | - 58.2 | - 38.8 |
| 34.7 | 56.3 | 36.9 | 45.4 | 36.3 | |||||||||
| SoCo | 100 | 40.4 | 60.4 | 43.7 | 34.9 | 56.8 | 37.0 | 41.1 | 61.0 | 44.4 | 35.6 | 57.5 | 38.0 |
| SoCo | 400 | 40.9 | 60.9 | 44.3 | 35.3 | 57.5 | 37.3 | 42.0 | 61.8 | 45.6 | 36.3 | 58.5 | 38.8 |
| (a) Sub-table 1. | ||||
| Methods | Epoch | APbb | AP | AP |
| Scratch Supervised | - 90 | 33.8 53.5 | 60.2 81.3 | 33.1 58.8 |
| ReSim-C4 [45] PixPro [10] InsLoc [12] DenseCL[11] | 200 400 400 200 | 58.7 60.2 58.4 58.7 | 83.1 83.8 83.0 82.8 | 66.3 67.7 65.3 65.2 |
| SoCo SoCo | 100 400 | 59.1 59.7 | 83.4 83.8 | 65.6 66.8 |
| Methods | Epoch | Apbb | AP | AP |
| MoCo[4] | 200 | 55.9 | 81.5 | 62.6 |
| SimCLR [9] | 1000 | 56.3 | 81.9 | 62.5 |
| MoCo v2[5] | 800 | 57.6 | 82.7 | 64.4 |
| InfoMin [6] | 200 | 57.6 | 82.7 | 64.6 |
| BYOL[3] | 300 | 51.9 | 81.0 | 56.5 |
| SwAV[7] | 400 | 45.1 | 77.4 | 46.5 |
| SimSiam[8] | 200 | 57.0 | 82.4 | 63.7 |
| ReSim-FPN[45] | 200 | 59.2 | 82.9 | 65.9 |
| Methods | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | Apmk | AP | APP | Apbb | AP8 | AP | Apmk | AP | APP | ||
| Scratch | - | 26.4 | 44.0 | 27.8 | 29.3 | 46.9 | 30.8 | 35.6 | 54.6 | 38.2 | 31.4 | 51.5 | 33.5 |
| Supervised | 90 | 38.2 | 58.2 | 41.2 | 33.3 | 54.7 | 35.2 | 40.0 | 59.9 | 43.1 | 34.7 | 56.5 | 36.9 |
| MoCo [4] | 200 | 38.5 | 58.3 | 41.6 | 33.6 | 54.8 | 35.6 | 40.7 | 60.5 | 44.1 | 35.4 | 57.3 | 37.6 37.1 |
| SimCLR [9] MoCo v2 [5] | 200 800 | - 39.3 | - | ■ | - | - | 1 | 39.6 41.2 | 59.1 60.9 | 42.9 44.6 | 34.6 35.8 | 55.9 57.7 | 38.2 |
| InfoMin 6] | 58.9 | 42.5 | 34.3 | 55.7 | 36.5 | 61.2 | 45.0 | 57.9 | 38.3 | ||||
| BYOL[3] | 200 300 | 39.0 | 58.5 | 42.0 | 34.1 | 55.2 | 36.3 | 41.3 | 36.0 | 56.8 | 37.3 | ||
| SwAV[7] | 400 | - | - | - | - | - | - | 40.3 | 60.5 | 43.9 | 35.1 | 36.6 | |
| SimSiam [8] | 200 | - 39.2 | - 59.3 | - | 1 | 1 | - | 39.6 | 60.1 | 42.9 | 34.7 | 56.6 | |
| PixPro[10] | 400 | 40.5 | 59.8 | 42.1 | 34.4 | 56.0 | 36.7 | - | ： | - | - | - | - |
| InsLoc [12] | 400 | 39.8 | 59.6 | 44.0 42.9 | - | - | - | - 41.8 | - 61.6 | - | - | - 58.2 | - 38.8 |
| 34.7 | 56.3 | 36.9 | 45.4 | 36.3 | |||||||||
| SoCo | 100 | 40.4 | 60.4 | 43.7 | 34.9 | 56.8 | 37.0 | 41.1 | 61.0 | 44.4 | 35.6 | 57.5 | 38.0 |
| SoCo | 400 | 40.9 | 60.9 | 44.3 | 35.3 | 57.5 | 37.3 | 42.0 | 61.8 | 45.6 | 36.3 | 58.5 | 38.8 |
| (a) Sub-table 1. | ||||
| Methods | Epoch | APbb | AP | AP |
| Scratch Supervised | - 90 | 33.8 53.5 | 60.2 81.3 | 33.1 58.8 |
| ReSim-C4 [45] PixPro [10] InsLoc [12] DenseCL[11] | 200 400 400 200 | 58.7 60.2 58.4 58.7 | 83.1 83.8 83.0 82.8 | 66.3 67.7 65.3 65.2 |
| SoCo SoCo | 100 400 | 59.1 59.7 | 83.4 83.8 | 65.6 66.8 |
| Methods | Epoch | Apbb | AP | AP |
| MoCo[4] | 200 | 55.9 | 81.5 | 62.6 |
| SimCLR [9] | 1000 | 56.3 | 81.9 | 62.5 |
| MoCo v2[5] | 800 | 57.6 | 82.7 | 64.4 |
| InfoMin [6] | 200 | 57.6 | 82.7 | 64.6 |
| BYOL[3] | 300 | 51.9 | 81.0 | 56.5 |
| SwAV[7] | 400 | 45.1 | 77.4 | 46.5 |
| SimSiam[8] | 200 | 57.0 | 82.4 | 63.7 |
| ReSim-FPN[45] | 200 | 59.2 | 82.9 | 65.9 |
| Whole Image | Selective Search | FPN | Head | Scale-aware Assignment | Box Jitter | Multi View | Apbb | Apmk |
| √ | 38.1 | 34.4 | ||||||
| √ | 40.6 (+2.5) | 36.8 (+2.4) | ||||||
| √ | 40.2 (+2.1) | 36.2 (+1.8) | ||||||
| √ | 广 | 41.2 (+3.1) | 37.0 (+2.6) | |||||
| √ | 1 | 41.6 (+3.5) | 37.3 (+2.9) | |||||
| √ | >>>>>> | >>>>> | 1 | 4 | 41.7 (+3.6) | 37.5 (+3.1) | ||
| √ | 厂 | 厂 | 42.3 (+4.2) | 37.6 (+3.2) |
| Image Size | Apbb | Apmk |
| 96 | 42.1 | 37.7 |
| 112 | 42.3 | 37.6 |
| 128 | 42.1 | 37.7 |
| 160 | 42.0 | 37.6 |
| 192 | 42.2 | 37.8 |
| Batch Size | APbb | Apmk |
| 512 | 41.7 | 37.6 |
| 1024 | 41.9 | 37.6 |
| 2048 | 42.3 | 37.6 |
| 4096 | 41.4 | 37.3 |
| Selective Search | Random | K | APbb | Apmk |
| < | 1 | 41.6 | 37.3 | |
| 486 | 42.3 41.6 | 37.6 | ||
| √ | 37.4 | |||
| √ | 41.2 | 37.0 | ||
| >>> | 148 | 41.4 | 36.9 | |
| NaN NaN | NaN NaN |
| T | APbb | Apmk |
| 0.98 | 35.0 | 31.7 |
| 0.99 | 42.3 | 37.6 |
| 0.993 | 41.8 | 37.6 |
| Whole Image | Selective Search | FPN | Head | Scale-aware Assignment | Box Jitter | Multi View | Apbb | Apmk |
| √ | 38.1 | 34.4 | ||||||
| √ | 40.6 (+2.5) | 36.8 (+2.4) | ||||||
| √ | 40.2 (+2.1) | 36.2 (+1.8) | ||||||
| √ | 广 | 41.2 (+3.1) | 37.0 (+2.6) | |||||
| √ | 1 | 41.6 (+3.5) | 37.3 (+2.9) | |||||
| √ | >>>>>> | >>>>> | 1 | 4 | 41.7 (+3.6) | 37.5 (+3.1) | ||
| √ | 厂 | 厂 | 42.3 (+4.2) | 37.6 (+3.2) |
| Image Size | Apbb | Apmk |
| 96 | 42.1 | 37.7 |
| 112 | 42.3 | 37.6 |
| 128 | 42.1 | 37.7 |
| 160 | 42.0 | 37.6 |
| 192 | 42.2 | 37.8 |
| Batch Size | APbb | Apmk |
| 512 | 41.7 | 37.6 |
| 1024 | 41.9 | 37.6 |
| 2048 | 42.3 | 37.6 |
| 4096 | 41.4 | 37.3 |
| Selective Search | Random | K | APbb | Apmk |
| < | 1 | 41.6 | 37.3 | |
| 486 | 42.3 41.6 | 37.6 | ||
| √ | 37.4 | |||
| √ | 41.2 | 37.0 | ||
| >>> | 148 | 41.4 | 36.9 | |
| NaN NaN | NaN NaN |
| T | APbb | Apmk |
| 0.98 | 35.0 | 31.7 |
| 0.99 | 42.3 | 37.6 |
| 0.993 | 41.8 | 37.6 |
| Method | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | APmk | AP | AP7 | Apbb AP | AP | Apmk | AP | AP | |||
| Supervised | 90 | 20.4 | 32.9 | 21.7 | 19.4 | 30.6 | 20.5 | 23.4 | 36.9 | 24.9 | 22.3 | 34.7 | 23.5 |
| SoCo* | 400 | 26.3 | 41.2 | 27.8 | 25.0 | 38.5 | 26.8 | 28.3 | 43.5 | 30.7 | 26.9 | 41.1 | 28.7 |
| Method | Epoch | Apbb | AP8 | AP |
| Supervised | 90 | 36.3 | 55.3 | 38.6 |
| SoCo* 一 | 400 | 38.3 | 57.2 | 41.2 |
| Method | Epoch | Apbb | AP8 | AP9 |
| Supervised | 90 | 36.6 | 56.0 | 38.8 |
| SoCo* | 400 | 37.4 | 56.3 | 39.9 |
| Method | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | APmk | AP | AP7 | Apbb AP | AP | Apmk | AP | AP | |||
| Supervised | 90 | 20.4 | 32.9 | 21.7 | 19.4 | 30.6 | 20.5 | 23.4 | 36.9 | 24.9 | 22.3 | 34.7 | 23.5 |
| SoCo* | 400 | 26.3 | 41.2 | 27.8 | 25.0 | 38.5 | 26.8 | 28.3 | 43.5 | 30.7 | 26.9 | 41.1 | 28.7 |
| Method | Epoch | Apbb | AP8 | AP |
| Supervised | 90 | 36.3 | 55.3 | 38.6 |
| SoCo* 一 | 400 | 38.3 | 57.2 | 41.2 |
| Method | Epoch | Apbb | AP8 | AP9 |
| Supervised | 90 | 36.6 | 56.0 | 38.8 |
| SoCo* | 400 | 37.4 | 56.3 | 39.9 |
| Methods | Epoch | Mini COCO (5%) | Mini C0CO (10%) | ||||||||||
| Apbb AP8 | AP9 | APmk | APk | APm | Apbb | AP8 | AP | Apmk | AP APP | ||||
| Supervised | 90 | 19.4 | 36.6 | 18.6 | 18.3 | 33.5 | 17.9 | 24.7 | 43.1 | 25.3 | 22.9 | 40.0 | 23.4 |
| SoCo | 100 | 24.6 | 42.2 | 25.6 | 22.1 | 38.8 | 22.4 | 29.2 | 47.7 | 31.1 | 26.1 | 44.4 | 27.0 |
| SoCo | 400 | 26.0 | 43.2 | 27.5 | 22.9 | 40.0 | 23.4 | 30.4 | 48.6 | 32.5 | 26.8 | 45.2 | 27.9 |
| SoCo* | 400 | 26.8 | 45.0 | 28.3 | 23.8 | 41.4 | 24.2 | 31.1 | 49.9 | 33.4 | 27.6 | 46.4 | 28.9 |
| Pretraining dataset | Epoch | Apbb | AP | AP | Apmk | AP | APP |
| ImageNet | 100 | 42.3 | 62.5 | 46.5 | 37.6 | 59.1 | 40.5 |
| ImageNet-subset | 100 | 36.9 | 56.2 | 40.1 | 33.2 | 53.0 | 35.6 |
| COCO train set + unlabeled set | 100 | 37.3 | 56.5 | 40.6 | 33.5 | 53.4 | 36.0 |
| COCO train set + unlabeled set | 530 | 40.6 | 61.1 | 44.4 | 36.4 | 58.1 | 38.7 |
| Method | Epoch | APbb | APb 50 | AP | Apmk | AP 50 | APP |
| Supervised | 90 | 41.9 | 61.5 | 45.4 | 37.7 | 58.8 | 40.5 |
| SoCo* | 400 | 44.5 | 64.2 | 48.8 | 39.6 | 61.5 | 42.4 |
| Methods | Epoch | Mini COCO (5%) | Mini C0CO (10%) | ||||||||||
| Apbb AP8 | AP9 | APmk | APk | APm | Apbb | AP8 | AP | Apmk | AP APP | ||||
| Supervised | 90 | 19.4 | 36.6 | 18.6 | 18.3 | 33.5 | 17.9 | 24.7 | 43.1 | 25.3 | 22.9 | 40.0 | 23.4 |
| SoCo | 100 | 24.6 | 42.2 | 25.6 | 22.1 | 38.8 | 22.4 | 29.2 | 47.7 | 31.1 | 26.1 | 44.4 | 27.0 |
| SoCo | 400 | 26.0 | 43.2 | 27.5 | 22.9 | 40.0 | 23.4 | 30.4 | 48.6 | 32.5 | 26.8 | 45.2 | 27.9 |
| SoCo* | 400 | 26.8 | 45.0 | 28.3 | 23.8 | 41.4 | 24.2 | 31.1 | 49.9 | 33.4 | 27.6 | 46.4 | 28.9 |
| Pretraining dataset | Epoch | Apbb | AP | AP | Apmk | AP | APP |
| ImageNet | 100 | 42.3 | 62.5 | 46.5 | 37.6 | 59.1 | 40.5 |
| ImageNet-subset | 100 | 36.9 | 56.2 | 40.1 | 33.2 | 53.0 | 35.6 |
| COCO train set + unlabeled set | 100 | 37.3 | 56.5 | 40.6 | 33.5 | 53.4 | 36.0 |
| COCO train set + unlabeled set | 530 | 40.6 | 61.1 | 44.4 | 36.4 | 58.1 | 38.7 |
| Method | Epoch | APbb | APb 50 | AP | Apmk | AP 50 | APP |
| Supervised | 90 | 41.9 | 61.5 | 45.4 | 37.7 | 58.8 | 40.5 |
| SoCo* | 400 | 44.5 | 64.2 | 48.8 | 39.6 | 61.5 | 42.4 |
| Methods | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | Apmk | AP | AP | Apbb | AP AP | Apmk | AP | AP | |||
| Scratch | - | 31.0 | 49.5 | 33.2 | 28.5 | 46.8 | 30.4 | 38.4 | 57.5 | 42.0 | 34.7 | 54.8 | 37.2 |
| Supervised | 90 | 38.9 | 59.6 | 42.7 | 35.4 | 56.5 | 38.1 | 41.3 | 61.3 | 45.0 | 37.3 | 58.3 | 40.3 |
| MoCo[4] | 200 | 38.5 | 58.9 | 42.0 | 35.1 | 55.9 | 37.7 | 40.8 | 61.6 | 44.7 | 36.9 | 58.4 | 39.7 |
| MoCo v2[5] | 200 | 40.4 | 60.2 | 44.2 | 36.4 | 57.2 | 38.9 | 41.7 | 61.6 | 45.6 | 37.6 | 58.7 | 40.5 |
| InfoMin [6] | 200 | 40.6 | 60.6 | 44.6 | 36.7 | 57.7 | 39.4 | 42.5 | 62.7 | 46.8 | 38.4 | 59.7 | 41.4 |
| BYOL[3] | 300 | 40.4 | 61.6 | 44.1 | 37.2 | 58.8 | 39.8 | 42.3 | 62.6 | 46.2 | 38.3 | 59.6 | 41.1 |
| SwAV[7] | 400 | - | - | - | - | - | - | 42.3 | 62.8 | 46.3 | 38.2 | 60.0 | 41.0 |
| ReSim-FPNT [45] | 200 | 39.8 | 60.2 | 43.5 | 36.0 | 57.1 | 38.6 | 41.4 | 61.9 | 45.4 | 37.5 | 59.1 | 40.3 |
| PixPro[10] | 400 | 41.4 | 61.6 | 45.4 | 1 | - | 1 | - | - | - | - | 1 | - |
| InsLoc [12] | 400 | 42.0 | 62.3 | 45.8 | 37.6 | 59.0 | 40.5 | 43.3 | 63.6 | 47.3 | 38.8 | 60.9 | 41.7 |
| DenseCL[11] | 200 | 40.3 | 59.9 | 44.3 | 36.4 | 57.0 | 39.2 | 41.2 | 61.9 | 45.1 | 37.3 | 58.9 | 40.1 |
| DetCons [13] | 1000 | 41.8 | 1 | 37.4 | - | ： | 42.9 | 1 | - | 38.1 | ： | ||
| DetConB [13] | 1000 | 42.7 | - | 38.2 | 1 | ： | 43.4 | 1 | - | 38.7 | - | ： | |
| SoCo | 100 | 42.3 | 62.5 | 46.5 | 37.6 | 59.1 | 40.5 | 43.2 | 63.3 | 47.3 | 38.8 | 60.6 | 41.9 |
| SoCo | 400 | 43.0 | 63.3 | 47.1 | 38.2 | 60.2 | 41.0 | 44.0 | 64.0 | 48.4 | 39.0 | 61.3 | 41.7 |
| SoCo* | 400 | 43.2 | 63.5 | 47.4 | 38.4 | 60.2 | 41.4 | 44.3 | 64.6 | 48.9 | 39.6 | 61.8 | 42.5 |
| Methods | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | Apmk | AP | APP | Apbb | AP8 | AP | Apmk | AP | APP | ||
| Scratch | - | 26.4 | 44.0 | 27.8 | 29.3 | 46.9 | 30.8 | 35.6 | 54.6 | 38.2 | 31.4 | 51.5 | 33.5 |
| Supervised | 90 | 38.2 | 58.2 | 41.2 | 33.3 | 54.7 | 35.2 | 40.0 | 59.9 | 43.1 | 34.7 | 56.5 | 36.9 |
| MoCo [4] | 200 | 38.5 | 58.3 | 41.6 | 33.6 | 54.8 | 35.6 | 40.7 | 60.5 | 44.1 | 35.4 | 57.3 | 37.6 37.1 |
| SimCLR [9] MoCo v2 [5] | 200 800 | - 39.3 | - | ■ | - | - | 1 | 39.6 41.2 | 59.1 60.9 | 42.9 44.6 | 34.6 35.8 | 55.9 57.7 | 38.2 |
| InfoMin 6] | 58.9 | 42.5 | 34.3 | 55.7 | 36.5 | 61.2 | 45.0 | 57.9 | 38.3 | ||||
| BYOL[3] | 200 300 | 39.0 | 58.5 | 42.0 | 34.1 | 55.2 | 36.3 | 41.3 | 36.0 | 56.8 | 37.3 | ||
| SwAV[7] | 400 | - | - | - | - | - | - | 40.3 | 60.5 | 43.9 | 35.1 | 36.6 | |
| SimSiam [8] | 200 | - 39.2 | - 59.3 | - | 1 | 1 | - | 39.6 | 60.1 | 42.9 | 34.7 | 56.6 | |
| PixPro[10] | 400 | 40.5 | 59.8 | 42.1 | 34.4 | 56.0 | 36.7 | - | ： | - | - | - | - |
| InsLoc [12] | 400 | 39.8 | 59.6 | 44.0 42.9 | - | - | - | - 41.8 | - 61.6 | - | - | - 58.2 | - 38.8 |
| 34.7 | 56.3 | 36.9 | 45.4 | 36.3 | |||||||||
| SoCo | 100 | 40.4 | 60.4 | 43.7 | 34.9 | 56.8 | 37.0 | 41.1 | 61.0 | 44.4 | 35.6 | 57.5 | 38.0 |
| SoCo | 400 | 40.9 | 60.9 | 44.3 | 35.3 | 57.5 | 37.3 | 42.0 | 61.8 | 45.6 | 36.3 | 58.5 | 38.8 |
| (a) Sub-table 1. | ||||
| Methods | Epoch | APbb | AP | AP |
| Scratch Supervised | - 90 | 33.8 53.5 | 60.2 81.3 | 33.1 58.8 |
| ReSim-C4 [45] PixPro [10] InsLoc [12] DenseCL[11] | 200 400 400 200 | 58.7 60.2 58.4 58.7 | 83.1 83.8 83.0 82.8 | 66.3 67.7 65.3 65.2 |
| SoCo SoCo | 100 400 | 59.1 59.7 | 83.4 83.8 | 65.6 66.8 |
| Methods | Epoch | Apbb | AP | AP |
| MoCo[4] | 200 | 55.9 | 81.5 | 62.6 |
| SimCLR [9] | 1000 | 56.3 | 81.9 | 62.5 |
| MoCo v2[5] | 800 | 57.6 | 82.7 | 64.4 |
| InfoMin [6] | 200 | 57.6 | 82.7 | 64.6 |
| BYOL[3] | 300 | 51.9 | 81.0 | 56.5 |
| SwAV[7] | 400 | 45.1 | 77.4 | 46.5 |
| SimSiam[8] | 200 | 57.0 | 82.4 | 63.7 |
| ReSim-FPN[45] | 200 | 59.2 | 82.9 | 65.9 |
| Whole Image | Selective Search | FPN | Head | Scale-aware Assignment | Box Jitter | Multi View | Apbb | Apmk |
| √ | 38.1 | 34.4 | ||||||
| √ | 40.6 (+2.5) | 36.8 (+2.4) | ||||||
| √ | 40.2 (+2.1) | 36.2 (+1.8) | ||||||
| √ | 广 | 41.2 (+3.1) | 37.0 (+2.6) | |||||
| √ | 1 | 41.6 (+3.5) | 37.3 (+2.9) | |||||
| √ | >>>>>> | >>>>> | 1 | 4 | 41.7 (+3.6) | 37.5 (+3.1) | ||
| √ | 厂 | 厂 | 42.3 (+4.2) | 37.6 (+3.2) |
| Selective Search | Random | K | APbb | Apmk |
| < | 1 | 41.6 | 37.3 | |
| 486 | 42.3 41.6 | 37.6 | ||
| √ | 37.4 | |||
| √ | 41.2 | 37.0 | ||
| >>> | 148 | 41.4 | 36.9 | |
| NaN NaN | NaN NaN |
| Image Size | Apbb | Apmk |
| 96 | 42.1 | 37.7 |
| 112 | 42.3 | 37.6 |
| 128 | 42.1 | 37.7 |
| 160 | 42.0 | 37.6 |
| 192 | 42.2 | 37.8 |
| Batch Size | APbb | Apmk |
| 512 | 41.7 | 37.6 |
| 1024 | 41.9 | 37.6 |
| 2048 | 42.3 | 37.6 |
| 4096 | 41.4 | 37.3 |
| T | APbb | Apmk |
| 0.98 | 35.0 | 31.7 |
| 0.99 | 42.3 | 37.6 |
| 0.993 | 41.8 | 37.6 |
| Method | Epoch | Apbb | AP8 | AP |
| Supervised | 90 | 36.3 | 55.3 | 38.6 |
| SoCo* 一 | 400 | 38.3 | 57.2 | 41.2 |
| Method | Epoch | 1× Schedule | 2× Schedule | ||||||||||
| Apbb | AP | AP | APmk | AP | AP7 | Apbb AP | AP | Apmk | AP | AP | |||
| Supervised | 90 | 20.4 | 32.9 | 21.7 | 19.4 | 30.6 | 20.5 | 23.4 | 36.9 | 24.9 | 22.3 | 34.7 | 23.5 |
| SoCo* | 400 | 26.3 | 41.2 | 27.8 | 25.0 | 38.5 | 26.8 | 28.3 | 43.5 | 30.7 | 26.9 | 41.1 | 28.7 |
| Method | Epoch | Apbb | AP8 | AP9 |
| Supervised | 90 | 36.6 | 56.0 | 38.8 |
| SoCo* | 400 | 37.4 | 56.3 | 39.9 |
| Pretraining dataset | Epoch | Apbb | AP | AP | Apmk | AP | APP |
| ImageNet | 100 | 42.3 | 62.5 | 46.5 | 37.6 | 59.1 | 40.5 |
| ImageNet-subset | 100 | 36.9 | 56.2 | 40.1 | 33.2 | 53.0 | 35.6 |
| COCO train set + unlabeled set | 100 | 37.3 | 56.5 | 40.6 | 33.5 | 53.4 | 36.0 |
| COCO train set + unlabeled set | 530 | 40.6 | 61.1 | 44.4 | 36.4 | 58.1 | 38.7 |
| Method | Epoch | APbb | APb 50 | AP | Apmk | AP 50 | APP |
| Supervised | 90 | 41.9 | 61.5 | 45.4 | 37.7 | 58.8 | 40.5 |
| SoCo* | 400 | 44.5 | 64.2 | 48.8 | 39.6 | 61.5 | 42.4 |
| Methods | Epoch | Mini COCO (5%) | Mini C0CO (10%) | ||||||||||
| Apbb AP8 | AP9 | APmk | APk | APm | Apbb | AP8 | AP | Apmk | AP APP | ||||
| Supervised | 90 | 19.4 | 36.6 | 18.6 | 18.3 | 33.5 | 17.9 | 24.7 | 43.1 | 25.3 | 22.9 | 40.0 | 23.4 |
| SoCo | 100 | 24.6 | 42.2 | 25.6 | 22.1 | 38.8 | 22.4 | 29.2 | 47.7 | 31.1 | 26.1 | 44.4 | 27.0 |
| SoCo | 400 | 26.0 | 43.2 | 27.5 | 22.9 | 40.0 | 23.4 | 30.4 | 48.6 | 32.5 | 26.8 | 45.2 | 27.9 |
| SoCo* | 400 | 26.8 | 45.0 | 28.3 | 23.8 | 41.4 | 24.2 | 31.1 | 49.9 | 33.4 | 27.6 | 46.4 | 28.9 |
| Num training envs: | 1 | 10 | 50 | 100 | 500 |
| PPO | |||||
| RGB | 0.21 ± 0.04 | 0.17± 0.04 | 0.35 ± 0.13 | 0.35 ± 0.16 | 0.34 ± 0.14 |
| RGB-D | 0.05±0.04 | 0.89 ±0.05 | 0.90±0.05 | 0.61 ±0.37 | 0.77 ± 0.33 |
| Grayscale | 0.36 ± 0.04 | 0.33 ± 0.13 | 0.37± 0.04 | 0.47±0.14 | 0.41 ± 0.22 |
| PPO-IR (split) | |||||
| RGB | 0.64 ± 0.05 | 0.69 ±0.03 | 0.72 ± 0.016 | 0.75 ± 0.02 | |
| RGB-D | 1 | 0.85 ± 0.02 | 0.90±0.05 | 0.94±0.01 | 0.95± 0.02 |
| Grayscale | 1 | 0.69±0.01 | 0.76±0.02 | 0.75 ±0.02 | 0.76±0.02 |
| PPO-IR (full objective) | |||||
| RGB | 1 | 0.79± 0.05 | 0.79 ±0.03 | 0.81 ±0.03 | 0.81±0.02 |
| RGB-D | 1 | 0.98±0.01 | 0.97 ± 0.01 | 0.99 ± 0.01 | 0.99± 0.01 |
| Grayscale | 1 | 0.79± 0.03 | 0.79±0.01 | 0.79±0.02 | 0.80±0.02 |
| Num training envs: | 1 | 10 | 50 | 100 | 500 |
| PPO | |||||
| RGB | 0.21 ± 0.04 | 0.17± 0.04 | 0.35 ± 0.13 | 0.35 ± 0.16 | 0.34 ± 0.14 |
| RGB-D | 0.05±0.04 | 0.89 ±0.05 | 0.90±0.05 | 0.61 ±0.37 | 0.77 ± 0.33 |
| Grayscale | 0.36 ± 0.04 | 0.33 ± 0.13 | 0.37± 0.04 | 0.47±0.14 | 0.41 ± 0.22 |
| PPO-IR (split) | |||||
| RGB | 0.64 ± 0.05 | 0.69 ±0.03 | 0.72 ± 0.016 | 0.75 ± 0.02 | |
| RGB-D | 1 | 0.85 ± 0.02 | 0.90±0.05 | 0.94±0.01 | 0.95± 0.02 |
| Grayscale | 1 | 0.69±0.01 | 0.76±0.02 | 0.75 ±0.02 | 0.76±0.02 |
| PPO-IR (full objective) | |||||
| RGB | 1 | 0.79± 0.05 | 0.79 ±0.03 | 0.81 ±0.03 | 0.81±0.02 |
| RGB-D | 1 | 0.98±0.01 | 0.97 ± 0.01 | 0.99 ± 0.01 | 0.99± 0.01 |
| Grayscale | 1 | 0.79± 0.03 | 0.79±0.01 | 0.79±0.02 | 0.80±0.02 |
| Initialize k1, k2,0o,Ti={..,N}, env while not converged do fori=1...,k1 do // Train πθ on env on the RL objective |
| 0 ←maxθ Lppo(oenv;πi-1) end for |
| for j=1...,k2 do |
| // Train π on env andT(enu) Sample {oenv,T01 （oenu）} |
| Generate {oTi(eno), (Ti(eno)}i=-. ,T0k1 |
| θj ← minθ d(π0κ (Oenv)llπθj-1(Ti(Oenv)); |
| end for end while |
| Layer | Quantization | cmul | Cadd | s | S |
| Fully connected | uniform pow-2 | MMl-1 0 | MM-1 2MMl-1 | M(M-1+1)b | Mb |
| Convolutional | uniform pow-2 | MM-1N²K² 0 | MM-1N²K² 2MM1-1NK2 | Mt(M-1K²+1)b | MN² |
| Bitwidth Weight/Activ. | qmax Weight/Activ. | Size Weight/Activ.(max)/Activ.(sum) | Validation error | Uniform quant. Power-of-two quant. Validation error | |
| Baseline | 32bit/32bit | 1 | 1048KB/64KB/736KB | 7.29% | |
| Fixed | 2bit/32bit | fixed/- | 65.5KB/64KB/736KB | 10.81% | 8.99% |
| TQT (Jain et al.,2019) | 2bit/32bit | learned/- | 65.5KB/64KB/736KB | 9.47% | 8.79% |
| Ours (w/constr. (8a)) | learned/32bit | learned/- | 70KB/64KB/736KB | 8.59% | 8.53% |
| Fixed | 2bit/4bit | fixed/fixed | 65.5KB/8KB/92KB | 11.30% | 11.62% |
| TQT (Jain et al.,2019) | 2bit/4bit | learned/learned | 65.5KB/8KB/92KB | 9.62% | 11.29% |
| Ours (w/constr.(8a) and(8b)) learned/learned learned/learned | 70KB/- /92KB | 9.38% | 11.29% | ||
| Ours (w/constr.(8a) and (8c)) | ）learned/learned | learned/learned | 70KB/8KB/- | 8.58% | 11.23% |
| Parametrization | SGD momentum | ADAM |
| U1 | 11.74 | 7.61 |
| U2 | 7.44 | 7.85 |
| U3 | 7.32 | 7.36 |
| P1 | 15.35 | 7.54 |
| P2 | 7.74 | 7.79 |
| P3 | 7.40 | 7.40 |
| Layer | Quantization | cmul | Cadd | s | S |
| Fully connected | uniform pow-2 | MMl-1 0 | MM-1 2MMl-1 | M(M-1+1)b | Mb |
| Convolutional | uniform pow-2 | MM-1N²K² 0 | MM-1N²K² 2MM1-1NK2 | Mt(M-1K²+1)b | MN² |
| Bitwidth Weight/Activ. | qmax Weight/Activ. | Size Weight/Activ.(max)/Activ.(sum) | Validation error | Uniform quant. Power-of-two quant. Validation error | |
| Baseline | 32bit/32bit | 1 | 1048KB/64KB/736KB | 7.29% | |
| Fixed | 2bit/32bit | fixed/- | 65.5KB/64KB/736KB | 10.81% | 8.99% |
| TQT (Jain et al.,2019) | 2bit/32bit | learned/- | 65.5KB/64KB/736KB | 9.47% | 8.79% |
| Ours (w/constr. (8a)) | learned/32bit | learned/- | 70KB/64KB/736KB | 8.59% | 8.53% |
| Fixed | 2bit/4bit | fixed/fixed | 65.5KB/8KB/92KB | 11.30% | 11.62% |
| TQT (Jain et al.,2019) | 2bit/4bit | learned/learned | 65.5KB/8KB/92KB | 9.62% | 11.29% |
| Ours (w/constr.(8a) and(8b)) learned/learned learned/learned | 70KB/- /92KB | 9.38% | 11.29% | ||
| Ours (w/constr.(8a) and (8c)) | ）learned/learned | learned/learned | 70KB/8KB/- | 8.58% | 11.23% |
| Parametrization | SGD momentum | ADAM |
| U1 | 11.74 | 7.61 |
| U2 | 7.44 | 7.85 |
| U3 | 7.32 | 7.36 |
| P1 | 15.35 | 7.54 |
| P2 | 7.74 | 7.79 |
| P3 | 7.40 | 7.40 |
| Layer | Quantization | cmul | Cadd | s | S |
| Fully connected | uniform pow-2 | MMl-1 0 | MM-1 2MMl-1 | M(M-1+1)b | Mb |
| Convolutional | uniform pow-2 | MM-1N²K² 0 | MM-1N²K² 2MM1-1NK2 | Mt(M-1K²+1)b | MN² |
| Layer | Quantization | cmul | Cadd | s | S |
| Fully connected | uniform pow-2 | MMl-1 0 | MM-1 2MMl-1 | M(M-1+1)b | Mb |
| Convolutional | uniform pow-2 | MM-1N²K² 0 | MM-1N²K² 2MM1-1NK2 | Mt(M-1K²+1)b | MN² |
| Bitwidth Weight/Activ. | qmax Weight/Activ. | Size Weight/Activ.(max)/Activ.(sum) | Validation error | Uniform quant. Power-of-two quant. Validation error | |
| Baseline | 32bit/32bit | 1 | 1048KB/64KB/736KB | 7.29% | |
| Fixed | 2bit/32bit | fixed/- | 65.5KB/64KB/736KB | 10.81% | 8.99% |
| TQT (Jain et al.,2019) | 2bit/32bit | learned/- | 65.5KB/64KB/736KB | 9.47% | 8.79% |
| Ours (w/constr. (8a)) | learned/32bit | learned/- | 70KB/64KB/736KB | 8.59% | 8.53% |
| Fixed | 2bit/4bit | fixed/fixed | 65.5KB/8KB/92KB | 11.30% | 11.62% |
| TQT (Jain et al.,2019) | 2bit/4bit | learned/learned | 65.5KB/8KB/92KB | 9.62% | 11.29% |
| Ours (w/constr.(8a) and(8b)) learned/learned learned/learned | 70KB/- /92KB | 9.38% | 11.29% | ||
| Ours (w/constr.(8a) and (8c)) | ）learned/learned | learned/learned | 70KB/8KB/- | 8.58% | 11.23% |
| Bitwidth Weight/Activ. | qmax Weight/Activ. | Size Weight/Activ.(max)/Activ.(sum) | Validation error | Uniform quant. Power-of-two quant. Validation error | |
| Baseline | 32bit/32bit | 1 | 1048KB/64KB/736KB | 7.29% | |
| Fixed | 2bit/32bit | fixed/- | 65.5KB/64KB/736KB | 10.81% | 8.99% |
| TQT (Jain et al.,2019) | 2bit/32bit | learned/- | 65.5KB/64KB/736KB | 9.47% | 8.79% |
| Ours (w/constr. (8a)) | learned/32bit | learned/- | 70KB/64KB/736KB | 8.59% | 8.53% |
| Fixed | 2bit/4bit | fixed/fixed | 65.5KB/8KB/92KB | 11.30% | 11.62% |
| TQT (Jain et al.,2019) | 2bit/4bit | learned/learned | 65.5KB/8KB/92KB | 9.62% | 11.29% |
| Ours (w/constr.(8a) and(8b)) learned/learned learned/learned | 70KB/- /92KB | 9.38% | 11.29% | ||
| Ours (w/constr.(8a) and (8c)) | ）learned/learned | learned/learned | 70KB/8KB/- | 8.58% | 11.23% |
| Parametrization | SGD momentum | ADAM |
| U1 | 11.74 | 7.61 |
| U2 | 7.44 | 7.85 |
| U3 | 7.32 | 7.36 |
| P1 | 15.35 | 7.54 |
| P2 | 7.74 | 7.79 |
| P3 | 7.40 | 7.40 |
| Parametrization | SGD momentum | ADAM |
| U1 | 11.74 | 7.61 |
| U2 | 7.44 | 7.85 |
| U3 | 7.32 | 7.36 |
| P1 | 15.35 | 7.54 |
| P2 | 7.74 | 7.79 |
| P3 | 7.40 | 7.40 |
| Quantization | Float32 | Uniform quantization | Power-of-two quantization | ||||||||||||
| θ=[b,d]T | 0=[b,qmax] | Θ= ：[d,qmax] T | 0=[b,qmax] T | 0=[b,qmin] T Θ= | [qmin， qmax] T | ||||||||||
| Weights Weights+Activations | 17.8%/8.18% 8.50%/7.29% 28.9%/9.03% | 8.80%/7.44% | 8.50%/7.32% 9.23%/7.40% | 11.70%/7.90% | 53.07%/23.01% | 10.61%/7.56% 15.10%/9.86% | |||||||||
| 10 | (a) Random initialization | 9.43%/7.74% | 22.91%/11.68% diverging/35.68% | ||||||||||||
| r Brrier 10 2 0 | 100 | 10 -1 | (b)Pre-trained initialization | 100 | 8 书 4 6 ·104 Iteration | ||||||||||
| 1 | 100 | ||||||||||||||
| 100 | |||||||||||||||
| Jl nriei 福 | 目 | ||||||||||||||
| 10 -1 | 10 2 1 | 10 1 | |||||||||||||
| 2 4 Iteration | 6 | 0 | 2 4 | 6 | 0 | 2 4 Iteration | 6 | 0 2 | |||||||
| Iteration b,d (U1) | ·104 | b,qmax (U2) d, qmax (U3) |
| Layer | Quantization | cmul | Cadd | s | S |
| Fully connected | uniform pow-2 | MMl-1 0 | MM-1 2MMl-1 | M(M-1+1)b | Mb |
| Convolutional | uniform pow-2 | MM-1N²K² 0 | MM-1N²K² 2MM1-1NK2 | Mt(M-1K²+1)b | MN² |
| Bitwidth Weight/Activ. | qmax Weight/Activ. | Size Weight/Activ.(max)/Activ.(sum) | Validation error | Uniform quant. Power-of-two quant. Validation error | |
| Baseline | 32bit/32bit | 1 | 1048KB/64KB/736KB | 7.29% | |
| Fixed | 2bit/32bit | fixed/- | 65.5KB/64KB/736KB | 10.81% | 8.99% |
| TQT (Jain et al.,2019) | 2bit/32bit | learned/- | 65.5KB/64KB/736KB | 9.47% | 8.79% |
| Ours (w/constr. (8a)) | learned/32bit | learned/- | 70KB/64KB/736KB | 8.59% | 8.53% |
| Fixed | 2bit/4bit | fixed/fixed | 65.5KB/8KB/92KB | 11.30% | 11.62% |
| TQT (Jain et al.,2019) | 2bit/4bit | learned/learned | 65.5KB/8KB/92KB | 9.62% | 11.29% |
| Ours (w/constr.(8a) and(8b)) learned/learned learned/learned | 70KB/- /92KB | 9.38% | 11.29% | ||
| Ours (w/constr.(8a) and (8c)) | ）learned/learned | learned/learned | 70KB/8KB/- | 8.58% | 11.23% |
| Parametrization | SGD momentum | ADAM |
| U1 | 11.74 | 7.61 |
| U2 | 7.44 | 7.85 |
| U3 | 7.32 | 7.36 |
| P1 | 15.35 | 7.54 |
| P2 | 7.74 | 7.79 |
| P3 | 7.40 | 7.40 |
| Model | In- distribution | fine-tuned test accuracy | Out- distribution | Mahalanobis AUROC | MSP AUROC |
| BiT-MR50x1 | CIFAR-100 | 87.01% | CIFAR-10 | 81.71% | 81.15% |
| BiT-MR101x3 | CIFAR-100 | 91.55% | CIFAR-10 | 90.10% | 83.69% |
| ViT-B_16 | CIFAR-100 | 90.95% | CIFAR-10 | 95.53% | 91.89% |
| R50+ViT-B_16 | CIFAR-100 | 91.71% | CIFAR-10 | 96.23% | 92.08% |
| MLP-Mixer-B_16 | CIFAR-100 | 90.40% | CIFAR-10 | 95.31% | 90.22% |
| BiT-MR50x1 BiT-MR101x3 | CIFAR-10 | 97.47% | CIFAR-100 | 95.52% | 85.87% |
| CIFAR-10 | 97.36% | CIFAR-100 | 94.55% | 85.34% | |
| ViT-B_16 | CIFAR-10 | 98.10% | CIFAR-100 | 98.42% | 97.68% |
| R50+ViT-B_16 | CIFAR-10 | 98.70% | CIFAR-100 | 98.52% | 97.75% |
| MLP-Mixer-B_16 | CIFAR-10 | 97.58% | CIFAR-100 | 97.85% | 96.28% |
| Model | In- distribution | fine-tuned test accuracy | Out- distribution | Mahalanobis AUROC | MSP AUROC |
| DINO ViT-B_16* | CIFAR-100 | 88.95% | CIFAR-10 | 88.78% | 81.25% |
| ViT-B_16 (early stop) | CIFAR-100 | 88.71% | CIFAR-10 | 93.05% | 88.82% |
| ViT-B_16 | CIFAR-100 | 90.95% | CIFAR-10 | 95.53% | 91.89% |
| R50+ViT-B_16 | CIFAR-100 | 91.71% | CIFAR-10 | 96.23% | 92.08% |
| ViT-L_16 | CIFAR-100 | 94.73% | CIFAR-10 | 97.98% | 94.28% |
| ViT ensemble | CIFAR-100 | CIFAR-10 | 98.11% | 95.15% |
| Number of OOD(CIFAR-10)examplesper class | R+ViT(withoutfine-tuning) | R+ViTfine-tuned onCIFAR-100 |
| 12310100 | 88.73±1.08%92.94 ± 0.55%93.25 ± 0.59%95.73 ± 0.31%97.70±0.01% | 98.70±0.08%99.02 ± 0.15%99.16 ± 0.11%99.46 ± 0.01%99.67±(0.01% |
| Number of OOD(CIFAR-100)examplesper class | R+ViT(withoutfine-tuning) | R+ViTfine-tuned onCIFAR-10 |
| 1 | 94.35± 0.05%95.10 ± 0.30%95.60 ± 0.01%96.42 ± 0.02%97.38±0.01% | 98.96±0.05%99.11 ± 0.04%99.17 ± 0.03%99.29 ± 0.02%99.50±(0.01% |
| 2 | ||
| 310 | ||
| 100 |
| Model | Test Accuracy | Mahalanobis AUROC | MSP AUROC |
| 1D CNN[Ren etal., 2019] | 85.93±0.11% | 64.75±0.73% | 65.84±0.46% |
| BERT pre-trainand fine-tune | 89.84±0.00% | 77.49±0.04% | 73.53±0.03% |
| Distribution 1 | Distribution 2 | Labels 1 | Labels 2 | AUROC |
| CIFAR-100CIFAR-100 | CIFAR-10CIFAR-10 | CIFAR-100 namesCIFAR-100 names | CIFAR-10 names | 69.49% |
| 94.68% | ||||
| CIFAR-10CIFAR-10 | CIFAR-100CIFAR-100 | CIFAR-10 namesCIFAR-10 names | CIFAR-100 names | 89.17%94.68% |
| CIFAR-100CIFAR-100 | SVHNSVHN | CIFAR-100 namesCIFAR-100 names | ["number"] | 93.05%99.67% |
| CIFAR-10CIFAR-10 | SVHNSVHN | CIFAR-10 namesCIFAR-10 names | ["number"] | 96.90%99.95% |
| Model | In- distribution | fine-tuned test accuracy | Out- distribution | Mahalanobis AUROC | MSP AUROC |
| BiT-MR50x1 | CIFAR-100 | 87.01% | CIFAR-10 | 81.71% | 81.15% |
| BiT-MR101x3 | CIFAR-100 | 91.55% | CIFAR-10 | 90.10% | 83.69% |
| ViT-B_16 | CIFAR-100 | 90.95% | CIFAR-10 | 95.53% | 91.89% |
| R50+ViT-B_16 | CIFAR-100 | 91.71% | CIFAR-10 | 96.23% | 92.08% |
| MLP-Mixer-B_16 | CIFAR-100 | 90.40% | CIFAR-10 | 95.31% | 90.22% |
| BiT-MR50x1 BiT-MR101x3 | CIFAR-10 | 97.47% | CIFAR-100 | 95.52% | 85.87% |
| CIFAR-10 | 97.36% | CIFAR-100 | 94.55% | 85.34% | |
| ViT-B_16 | CIFAR-10 | 98.10% | CIFAR-100 | 98.42% | 97.68% |
| R50+ViT-B_16 | CIFAR-10 | 98.70% | CIFAR-100 | 98.52% | 97.75% |
| MLP-Mixer-B_16 | CIFAR-10 | 97.58% | CIFAR-100 | 97.85% | 96.28% |
| Model | In- distribution | fine-tuned test accuracy | Out- distribution | Mahalanobis AUROC | MSP AUROC |
| DINO ViT-B_16* | CIFAR-100 | 88.95% | CIFAR-10 | 88.78% | 81.25% |
| ViT-B_16 (early stop) | CIFAR-100 | 88.71% | CIFAR-10 | 93.05% | 88.82% |
| ViT-B_16 | CIFAR-100 | 90.95% | CIFAR-10 | 95.53% | 91.89% |
| R50+ViT-B_16 | CIFAR-100 | 91.71% | CIFAR-10 | 96.23% | 92.08% |
| ViT-L_16 | CIFAR-100 | 94.73% | CIFAR-10 | 97.98% | 94.28% |
| ViT ensemble | CIFAR-100 | CIFAR-10 | 98.11% | 95.15% |
| Number of OOD(CIFAR-10)examplesper class | R+ViT(withoutfine-tuning) | R+ViTfine-tuned onCIFAR-100 |
| 12310100 | 88.73±1.08%92.94 ± 0.55%93.25 ± 0.59%95.73 ± 0.31%97.70±0.01% | 98.70±0.08%99.02 ± 0.15%99.16 ± 0.11%99.46 ± 0.01%99.67±(0.01% |
| Number of OOD(CIFAR-100)examplesper class | R+ViT(withoutfine-tuning) | R+ViTfine-tuned onCIFAR-10 |
| 1 | 94.35± 0.05%95.10 ± 0.30%95.60 ± 0.01%96.42 ± 0.02%97.38±0.01% | 98.96±0.05%99.11 ± 0.04%99.17 ± 0.03%99.29 ± 0.02%99.50±(0.01% |
| 2 | ||
| 310 | ||
| 100 |
| Model | Test Accuracy | Mahalanobis AUROC | MSP AUROC |
| 1D CNN[Ren etal., 2019] | 85.93±0.11% | 64.75±0.73% | 65.84±0.46% |
| BERT pre-trainand fine-tune | 89.84±0.00% | 77.49±0.04% | 73.53±0.03% |
| Distribution 1 | Distribution 2 | Labels 1 | Labels 2 | AUROC |
| CIFAR-100CIFAR-100 | CIFAR-10CIFAR-10 | CIFAR-100 namesCIFAR-100 names | CIFAR-10 names | 69.49% |
| 94.68% | ||||
| CIFAR-10CIFAR-10 | CIFAR-100CIFAR-100 | CIFAR-10 namesCIFAR-10 names | CIFAR-100 names | 89.17%94.68% |
| CIFAR-100CIFAR-100 | SVHNSVHN | CIFAR-100 namesCIFAR-100 names | ["number"] | 93.05%99.67% |
| CIFAR-10CIFAR-10 | SVHNSVHN | CIFAR-10 namesCIFAR-10 names | ["number"] | 96.90%99.95% |
| Model | In- distribution | fine-tuned test accuracy | Out- distribution | Mahalanobis AUROC | MSP AUROC |
| BiT-MR50x1 | CIFAR-100 | 87.01% | CIFAR-10 | 81.71% | 81.15% |
| BiT-MR101x3 | CIFAR-100 | 91.55% | CIFAR-10 | 90.10% | 83.69% |
| ViT-B_16 | CIFAR-100 | 90.95% | CIFAR-10 | 95.53% | 91.89% |
| R50+ViT-B_16 | CIFAR-100 | 91.71% | CIFAR-10 | 96.23% | 92.08% |
| MLP-Mixer-B_16 | CIFAR-100 | 90.40% | CIFAR-10 | 95.31% | 90.22% |
| BiT-MR50x1 BiT-MR101x3 | CIFAR-10 | 97.47% | CIFAR-100 | 95.52% | 85.87% |
| CIFAR-10 | 97.36% | CIFAR-100 | 94.55% | 85.34% | |
| ViT-B_16 | CIFAR-10 | 98.10% | CIFAR-100 | 98.42% | 97.68% |
| R50+ViT-B_16 | CIFAR-10 | 98.70% | CIFAR-100 | 98.52% | 97.75% |
| MLP-Mixer-B_16 | CIFAR-10 | 97.58% | CIFAR-100 | 97.85% | 96.28% |
| Model | In- distribution | fine-tuned test accuracy | Out- distribution | Mahalanobis AUROC | MSP AUROC |
| BiT-MR50x1 | CIFAR-100 | 87.01% | CIFAR-10 | 81.71% | 81.15% |
| BiT-MR101x3 | CIFAR-100 | 91.55% | CIFAR-10 | 90.10% | 83.69% |
| ViT-B_16 | CIFAR-100 | 90.95% | CIFAR-10 | 95.53% | 91.89% |
| R50+ViT-B_16 | CIFAR-100 | 91.71% | CIFAR-10 | 96.23% | 92.08% |
| MLP-Mixer-B_16 | CIFAR-100 | 90.40% | CIFAR-10 | 95.31% | 90.22% |
| BiT-MR50x1 BiT-MR101x3 | CIFAR-10 | 97.47% | CIFAR-100 | 95.52% | 85.87% |
| CIFAR-10 | 97.36% | CIFAR-100 | 94.55% | 85.34% | |
| ViT-B_16 | CIFAR-10 | 98.10% | CIFAR-100 | 98.42% | 97.68% |
| R50+ViT-B_16 | CIFAR-10 | 98.70% | CIFAR-100 | 98.52% | 97.75% |
| MLP-Mixer-B_16 | CIFAR-10 | 97.58% | CIFAR-100 | 97.85% | 96.28% |
| Model | In- distribution | fine-tuned test accuracy | Out- distribution | Mahalanobis AUROC | MSP AUROC |
| DINO ViT-B_16* | CIFAR-100 | 88.95% | CIFAR-10 | 88.78% | 81.25% |
| ViT-B_16 (early stop) | CIFAR-100 | 88.71% | CIFAR-10 | 93.05% | 88.82% |
| ViT-B_16 | CIFAR-100 | 90.95% | CIFAR-10 | 95.53% | 91.89% |
| R50+ViT-B_16 | CIFAR-100 | 91.71% | CIFAR-10 | 96.23% | 92.08% |
| ViT-L_16 | CIFAR-100 | 94.73% | CIFAR-10 | 97.98% | 94.28% |
| ViT ensemble | CIFAR-100 | CIFAR-10 | 98.11% | 95.15% |
| Model | In- distribution | fine-tuned test accuracy | Out- distribution | Mahalanobis AUROC | MSP AUROC |
| DINO ViT-B_16* | CIFAR-100 | 88.95% | CIFAR-10 | 88.78% | 81.25% |
| ViT-B_16 (early stop) | CIFAR-100 | 88.71% | CIFAR-10 | 93.05% | 88.82% |
| ViT-B_16 | CIFAR-100 | 90.95% | CIFAR-10 | 95.53% | 91.89% |
| R50+ViT-B_16 | CIFAR-100 | 91.71% | CIFAR-10 | 96.23% | 92.08% |
| ViT-L_16 | CIFAR-100 | 94.73% | CIFAR-10 | 97.98% | 94.28% |
| ViT ensemble | CIFAR-100 | CIFAR-10 | 98.11% | 95.15% |
| Number of OOD(CIFAR-10)examplesper class | R+ViT(withoutfine-tuning) | R+ViTfine-tuned onCIFAR-100 |
| 12310100 | 88.73±1.08%92.94 ± 0.55%93.25 ± 0.59%95.73 ± 0.31%97.70±0.01% | 98.70±0.08%99.02 ± 0.15%99.16 ± 0.11%99.46 ± 0.01%99.67±(0.01% |
| Number of OOD(CIFAR-100)examplesper class | R+ViT(withoutfine-tuning) | R+ViTfine-tuned onCIFAR-10 |
| 1 | 94.35± 0.05%95.10 ± 0.30%95.60 ± 0.01%96.42 ± 0.02%97.38±0.01% | 98.96±0.05%99.11 ± 0.04%99.17 ± 0.03%99.29 ± 0.02%99.50±(0.01% |
| 2 | ||
| 310 | ||
| 100 |
| Number of OOD(CIFAR-10)examplesper class | R+ViT(withoutfine-tuning) | R+ViTfine-tuned onCIFAR-100 |
| 12310100 | 88.73±1.08%92.94 ± 0.55%93.25 ± 0.59%95.73 ± 0.31%97.70±0.01% | 98.70±0.08%99.02 ± 0.15%99.16 ± 0.11%99.46 ± 0.01%99.67±(0.01% |
| Number of OOD(CIFAR-100)examplesper class | R+ViT(withoutfine-tuning) | R+ViTfine-tuned onCIFAR-10 |
| 1 | 94.35± 0.05%95.10 ± 0.30%95.60 ± 0.01%96.42 ± 0.02%97.38±0.01% | 98.96±0.05%99.11 ± 0.04%99.17 ± 0.03%99.29 ± 0.02%99.50±(0.01% |
| 2 | ||
| 310 | ||
| 100 |
| Model | Test Accuracy | Mahalanobis AUROC | MSP AUROC |
| 1D CNN[Ren etal., 2019] | 85.93±0.11% | 64.75±0.73% | 65.84±0.46% |
| BERT pre-trainand fine-tune | 89.84±0.00% | 77.49±0.04% | 73.53±0.03% |
| Model | Test Accuracy | Mahalanobis AUROC | MSP AUROC |
| 1D CNN[Ren etal., 2019] | 85.93±0.11% | 64.75±0.73% | 65.84±0.46% |
| BERT pre-trainand fine-tune | 89.84±0.00% | 77.49±0.04% | 73.53±0.03% |
| Distribution 1 | Distribution 2 | Labels 1 | Labels 2 | AUROC |
| CIFAR-100CIFAR-100 | CIFAR-10CIFAR-10 | CIFAR-100 namesCIFAR-100 names | CIFAR-10 names | 69.49% |
| 94.68% | ||||
| CIFAR-10CIFAR-10 | CIFAR-100CIFAR-100 | CIFAR-10 namesCIFAR-10 names | CIFAR-100 names | 89.17%94.68% |
| CIFAR-100CIFAR-100 | SVHNSVHN | CIFAR-100 namesCIFAR-100 names | ["number"] | 93.05%99.67% |
| CIFAR-10CIFAR-10 | SVHNSVHN | CIFAR-10 namesCIFAR-10 names | ["number"] | 96.90%99.95% |
| Distribution 1 | Distribution 2 | Labels 1 | Labels 2 | AUROC |
| CIFAR-100CIFAR-100 | CIFAR-10CIFAR-10 | CIFAR-100 namesCIFAR-100 names | CIFAR-10 names | 69.49% |
| 94.68% | ||||
| CIFAR-10CIFAR-10 | CIFAR-100CIFAR-100 | CIFAR-10 namesCIFAR-10 names | CIFAR-100 names | 89.17%94.68% |
| CIFAR-100CIFAR-100 | SVHNSVHN | CIFAR-100 namesCIFAR-100 names | ["number"] | 93.05%99.67% |
| CIFAR-10CIFAR-10 | SVHNSVHN | CIFAR-10 namesCIFAR-10 names | ["number"] | 96.90%99.95% |
| Model | In- distribution | fine-tuned test accuracy | Out- distribution | Mahalanobis AUROC | MSP AUROC |
| BiT-MR50x1 | CIFAR-100 | 87.01% | CIFAR-10 | 81.71% | 81.15% |
| BiT-MR101x3 | CIFAR-100 | 91.55% | CIFAR-10 | 90.10% | 83.69% |
| ViT-B_16 | CIFAR-100 | 90.95% | CIFAR-10 | 95.53% | 91.89% |
| R50+ViT-B_16 | CIFAR-100 | 91.71% | CIFAR-10 | 96.23% | 92.08% |
| MLP-Mixer-B_16 | CIFAR-100 | 90.40% | CIFAR-10 | 95.31% | 90.22% |
| BiT-MR50x1 BiT-MR101x3 | CIFAR-10 | 97.47% | CIFAR-100 | 95.52% | 85.87% |
| CIFAR-10 | 97.36% | CIFAR-100 | 94.55% | 85.34% | |
| ViT-B_16 | CIFAR-10 | 98.10% | CIFAR-100 | 98.42% | 97.68% |
| R50+ViT-B_16 | CIFAR-10 | 98.70% | CIFAR-100 | 98.52% | 97.75% |
| MLP-Mixer-B_16 | CIFAR-10 | 97.58% | CIFAR-100 | 97.85% | 96.28% |
| Model | In- distribution | fine-tuned test accuracy | Out- distribution | Mahalanobis AUROC | MSP AUROC |
| DINO ViT-B_16* | CIFAR-100 | 88.95% | CIFAR-10 | 88.78% | 81.25% |
| ViT-B_16 (early stop) | CIFAR-100 | 88.71% | CIFAR-10 | 93.05% | 88.82% |
| ViT-B_16 | CIFAR-100 | 90.95% | CIFAR-10 | 95.53% | 91.89% |
| R50+ViT-B_16 | CIFAR-100 | 91.71% | CIFAR-10 | 96.23% | 92.08% |
| ViT-L_16 | CIFAR-100 | 94.73% | CIFAR-10 | 97.98% | 94.28% |
| ViT ensemble | CIFAR-100 | CIFAR-10 | 98.11% | 95.15% |
| Number of OOD(CIFAR-100)examplesper class | R+ViT(withoutfine-tuning) | R+ViTfine-tuned onCIFAR-10 |
| 1 | 94.35± 0.05%95.10 ± 0.30%95.60 ± 0.01%96.42 ± 0.02%97.38±0.01% | 98.96±0.05%99.11 ± 0.04%99.17 ± 0.03%99.29 ± 0.02%99.50±(0.01% |
| 2 | ||
| 310 | ||
| 100 |
| Number of OOD(CIFAR-10)examplesper class | R+ViT(withoutfine-tuning) | R+ViTfine-tuned onCIFAR-100 |
| 12310100 | 88.73±1.08%92.94 ± 0.55%93.25 ± 0.59%95.73 ± 0.31%97.70±0.01% | 98.70±0.08%99.02 ± 0.15%99.16 ± 0.11%99.46 ± 0.01%99.67±(0.01% |
| Model | Test Accuracy | Mahalanobis AUROC | MSP AUROC |
| 1D CNN[Ren etal., 2019] | 85.93±0.11% | 64.75±0.73% | 65.84±0.46% |
| BERT pre-trainand fine-tune | 89.84±0.00% | 77.49±0.04% | 73.53±0.03% |
| Distribution 1 | Distribution 2 | Labels 1 | Labels 2 | AUROC |
| CIFAR-100CIFAR-100 | CIFAR-10CIFAR-10 | CIFAR-100 namesCIFAR-100 names | CIFAR-10 names | 69.49% |
| 94.68% | ||||
| CIFAR-10CIFAR-10 | CIFAR-100CIFAR-100 | CIFAR-10 namesCIFAR-10 names | CIFAR-100 names | 89.17%94.68% |
| CIFAR-100CIFAR-100 | SVHNSVHN | CIFAR-100 namesCIFAR-100 names | ["number"] | 93.05%99.67% |
| CIFAR-10CIFAR-10 | SVHNSVHN | CIFAR-10 namesCIFAR-10 names | ["number"] | 96.90%99.95% |
| Stag | Hare | |
| Stag | a,a | c,b |
| Hare | b,c | d,d |
| PBT | RR | RPG | RND | |
| Rew. | 3.8(0.3) | 3.8(0.2) | 4.3(0.2) | 2.8(0.3) |
| #Coop. | 1.9(0.2) | 2.2(0.1) | 2.0(0.3) | 1.3(0.2) |
| #Hunt | 0.6(0.1) | 0.4(0.0) | 0.7(0.0) | 0.6(0.1) |
| PBT | w=[0.5,1] | w=[0,1] | w=[0,0] | RPG | RND | |
| Rew. | 3.3(0.2) | 4.8(0.6) | 5.1(0.4) | 6.0(0.5) | 8.9(0.3) | 3.2(0.2) |
| #Attack | 0.4(0.0) | 0.7(0.2) | 0.3(0.1) | 0.5(0.1) | 0.9(0.1) | 0.4(0.0) |
| #Coop. | 0.0(0.0) | 0.6(0.6) | 2.3(0.3) | 1.6(0.1) | 2.0(0.2) | 0.0(0.0) |
| #Hunt | 0.7(0.1) | 0.6(0.3) | 0.3(0.0) | 0.7(0.0) | 0.9(0.1) | 0.7(0.0) |
| Agent | Agent | Adapt. | apt. | |||
| Opponent:Cooperative → Competitive | ||||||
| #Attack | 0.2(0.0) | 0.3(0.0) | 0.1(0.1) | |||
| Rew. | 0.7(0.7) | -0.2(0.6) | 0.8(0.5) | |||
| Opponent: Competitive→Cooperative | ||||||
| #Coop. | 1.0(0.3) | 1.4(0.4) | 0.3(0.4) | |||
| Rew. | 2.5(0.7) | 3.6(1.2) | 1.1(0.7) | |||
| Hyper-parameters | Value |
| Initial learning rate | 1e-3 |
| Minibatch size | 320 chunks of 10 timesteps |
| Adam stepsize (ε) | 1e-5 |
| Discount rate (γ) | 0.99 |
| GAE parameter (入) | 0.95 |
| Value loss coefficient | 1 |
| Entropy coefficient | 0.01 |
| Gradient clipping | 0.5 |
| PPO clipping parameter | 0.2 |
| Parallel threads | 64(Escalation),256(Monster-Hunt) |
| PPO epochs | 4 |
| reward scale parameter | 0.1 |
| episode length | 50 |
| Hyper-parameters | Value |
| Learning rate | 2.5e-4 |
| Minibatch size | 2 * 512 chunks of 32 timesteps |
| Adam stepsize (ε) | 1e-5 |
| Discount rate () | 0.995 |
| GAE parameter (入) | 0.95 |
| Value loss coefficient | 0.5 |
| action loss coefficient | 1 |
| Entropy coefficient | 0.01(discrete), 0.0025(continuous) |
| Gradient clipping | 20 |
| PPO clipping parameter | 0.1 |
| Parallel threads | 128 |
| PPO epochs | 4 |
| episode length | 128 |
| Settings | Policy | Rewards | #Split | #Hunt | #Attack | #Cooperate |
| Standard | w=[1,1] RPG | 3.843(0.23) 4.34(0.171) | 0.859(0.083) 0.971(0.13) | 0.411(0.034) 0.659(0.048) | 0.526(0.064) 0.548(0.038) | 2.203(0.136) 2.028(0.297) |
| w=[0.5,1] | 3.827(0.489) | 0.807(0.192) | 0.365(0.106) | 0.15(0.064) | 2.342(0.286) | |
| w=[1,0.5] | 3.174(0.653) | 0.718(0.148) | 0.432(0.026) | 0.458(0.031) | 1.716(0.418) | |
| Original | 1.08(0.836) | 0.3(0.19) | 0.361(0.134) | 0.291(0.098) | 0.483(0.442) | |
| RND PBT | 2.789(0.346) | 0.499(0.061) | 0.623(0.128) | 0.242(0.037) | 1.349(0.164) | |
| Aggressive | 3.822(0.347) | 0.744(0.129) | 0.585(0.146) | 0.297(0.055) | 1.935(0.167) | |
| w=[0,0] | 5.966(0.539) | 1.195(0.155) | 0.699(0.008) | 0.517(0.066) | 1.603(0.127) | |
| RPG | 8.907(0.292) | 1.655(0.138) | 0.862(0.053) | 0.903(0.081) | 2.039(0.209) | |
| w=[0,1] | 5.066(0.375) | 0.785(0.041) | 0.344(0.049) | 0.346(0.058) | 2.327(0.311) | |
| w=[1,1] | 4.622(0.277) | 0.836(0.304) | 0.934(0.108) | 0.552(0.019) | 0.028(0.023) | |
| w=[0.5,1] | 4.79(0.588) | 0.678(0.31) | 0.617(0.28) | 0.67(0.194) | 0.55(0.643) | |
| Original | 3.551(0.121) | 0.717(0.032) | 0.812(0.078) | 0.412(0.018) | 0.027(0.026) | |
| RND | 3.189(0.154) | 0.626(0.065) | 0.705(0.008) | 0.382(0.029) | 0.035(0.027) | |
| PBT | 3.348(0.222) | 0.697(0.133) | 0.732(0.096) | 0.396(0.014) | 0.007(0.005) |
| Original | 1w=[4,0,0,0]w=[0,0,0,4]w=[0,4,4,0]w=[4,1,4,0] | ||||
| #Rewards | 20.00(0.00) | 74.76(2.88) | 20.00(0.00) | -470.0(0.00) | -453.45(0.25) |
| Oppo. Type | Stag | Hare | TFT | Random |
| #Stag | 9.31(0.77) | 3.6(4.33) | 7.31(3.82) | 5.35(3.48) |
| #Hare | 0.69(0.77) | 6.4(4.33) | 2.69(3.81) | 4.65(3.48) |
| Stag | Hare | |
| Stag | a,a | c,b |
| Hare | b,c | d,d |
| PBT | RR | RPG | RND | |
| Rew. | 3.8(0.3) | 3.8(0.2) | 4.3(0.2) | 2.8(0.3) |
| #Coop. | 1.9(0.2) | 2.2(0.1) | 2.0(0.3) | 1.3(0.2) |
| #Hunt | 0.6(0.1) | 0.4(0.0) | 0.7(0.0) | 0.6(0.1) |
| PBT | w=[0.5,1] | w=[0,1] | w=[0,0] | RPG | RND | |
| Rew. | 3.3(0.2) | 4.8(0.6) | 5.1(0.4) | 6.0(0.5) | 8.9(0.3) | 3.2(0.2) |
| #Attack | 0.4(0.0) | 0.7(0.2) | 0.3(0.1) | 0.5(0.1) | 0.9(0.1) | 0.4(0.0) |
| #Coop. | 0.0(0.0) | 0.6(0.6) | 2.3(0.3) | 1.6(0.1) | 2.0(0.2) | 0.0(0.0) |
| #Hunt | 0.7(0.1) | 0.6(0.3) | 0.3(0.0) | 0.7(0.0) | 0.9(0.1) | 0.7(0.0) |
| Agent | Agent | Adapt. | apt. | |||
| Opponent:Cooperative → Competitive | ||||||
| #Attack | 0.2(0.0) | 0.3(0.0) | 0.1(0.1) | |||
| Rew. | 0.7(0.7) | -0.2(0.6) | 0.8(0.5) | |||
| Opponent: Competitive→Cooperative | ||||||
| #Coop. | 1.0(0.3) | 1.4(0.4) | 0.3(0.4) | |||
| Rew. | 2.5(0.7) | 3.6(1.2) | 1.1(0.7) | |||
| Hyper-parameters | Value |
| Initial learning rate | 1e-3 |
| Minibatch size | 320 chunks of 10 timesteps |
| Adam stepsize (ε) | 1e-5 |
| Discount rate (γ) | 0.99 |
| GAE parameter (入) | 0.95 |
| Value loss coefficient | 1 |
| Entropy coefficient | 0.01 |
| Gradient clipping | 0.5 |
| PPO clipping parameter | 0.2 |
| Parallel threads | 64(Escalation),256(Monster-Hunt) |
| PPO epochs | 4 |
| reward scale parameter | 0.1 |
| episode length | 50 |
| Hyper-parameters | Value |
| Learning rate | 2.5e-4 |
| Minibatch size | 2 * 512 chunks of 32 timesteps |
| Adam stepsize (ε) | 1e-5 |
| Discount rate () | 0.995 |
| GAE parameter (入) | 0.95 |
| Value loss coefficient | 0.5 |
| action loss coefficient | 1 |
| Entropy coefficient | 0.01(discrete), 0.0025(continuous) |
| Gradient clipping | 20 |
| PPO clipping parameter | 0.1 |
| Parallel threads | 128 |
| PPO epochs | 4 |
| episode length | 128 |
| Settings | Policy | Rewards | #Split | #Hunt | #Attack | #Cooperate |
| Standard | w=[1,1] RPG | 3.843(0.23) 4.34(0.171) | 0.859(0.083) 0.971(0.13) | 0.411(0.034) 0.659(0.048) | 0.526(0.064) 0.548(0.038) | 2.203(0.136) 2.028(0.297) |
| w=[0.5,1] | 3.827(0.489) | 0.807(0.192) | 0.365(0.106) | 0.15(0.064) | 2.342(0.286) | |
| w=[1,0.5] | 3.174(0.653) | 0.718(0.148) | 0.432(0.026) | 0.458(0.031) | 1.716(0.418) | |
| Original | 1.08(0.836) | 0.3(0.19) | 0.361(0.134) | 0.291(0.098) | 0.483(0.442) | |
| RND PBT | 2.789(0.346) | 0.499(0.061) | 0.623(0.128) | 0.242(0.037) | 1.349(0.164) | |
| Aggressive | 3.822(0.347) | 0.744(0.129) | 0.585(0.146) | 0.297(0.055) | 1.935(0.167) | |
| w=[0,0] | 5.966(0.539) | 1.195(0.155) | 0.699(0.008) | 0.517(0.066) | 1.603(0.127) | |
| RPG | 8.907(0.292) | 1.655(0.138) | 0.862(0.053) | 0.903(0.081) | 2.039(0.209) | |
| w=[0,1] | 5.066(0.375) | 0.785(0.041) | 0.344(0.049) | 0.346(0.058) | 2.327(0.311) | |
| w=[1,1] | 4.622(0.277) | 0.836(0.304) | 0.934(0.108) | 0.552(0.019) | 0.028(0.023) | |
| w=[0.5,1] | 4.79(0.588) | 0.678(0.31) | 0.617(0.28) | 0.67(0.194) | 0.55(0.643) | |
| Original | 3.551(0.121) | 0.717(0.032) | 0.812(0.078) | 0.412(0.018) | 0.027(0.026) | |
| RND | 3.189(0.154) | 0.626(0.065) | 0.705(0.008) | 0.382(0.029) | 0.035(0.027) | |
| PBT | 3.348(0.222) | 0.697(0.133) | 0.732(0.096) | 0.396(0.014) | 0.007(0.005) |
| Original | 1w=[4,0,0,0]w=[0,0,0,4]w=[0,4,4,0]w=[4,1,4,0] | ||||
| #Rewards | 20.00(0.00) | 74.76(2.88) | 20.00(0.00) | -470.0(0.00) | -453.45(0.25) |
| Oppo. Type | Stag | Hare | TFT | Random |
| #Stag | 9.31(0.77) | 3.6(4.33) | 7.31(3.82) | 5.35(3.48) |
| #Hare | 0.69(0.77) | 6.4(4.33) | 2.69(3.81) | 4.65(3.48) |
| Stag | Hare | |
| Stag | a,a | c,b |
| Hare | b,c | d,d |
| Stag | Hare | |
| Stag | a,a | c,b |
| Hare | b,c | d,d |
| PBT | RR | RPG | RND | |
| Rew. | 3.8(0.3) | 3.8(0.2) | 4.3(0.2) | 2.8(0.3) |
| #Coop. | 1.9(0.2) | 2.2(0.1) | 2.0(0.3) | 1.3(0.2) |
| #Hunt | 0.6(0.1) | 0.4(0.0) | 0.7(0.0) | 0.6(0.1) |
| PBT | RR | RPG | RND | |
| Rew. | 3.8(0.3) | 3.8(0.2) | 4.3(0.2) | 2.8(0.3) |
| #Coop. | 1.9(0.2) | 2.2(0.1) | 2.0(0.3) | 1.3(0.2) |
| #Hunt | 0.6(0.1) | 0.4(0.0) | 0.7(0.0) | 0.6(0.1) |
| PBT | w=[0.5,1] | w=[0,1] | w=[0,0] | RPG | RND | |
| Rew. | 3.3(0.2) | 4.8(0.6) | 5.1(0.4) | 6.0(0.5) | 8.9(0.3) | 3.2(0.2) |
| #Attack | 0.4(0.0) | 0.7(0.2) | 0.3(0.1) | 0.5(0.1) | 0.9(0.1) | 0.4(0.0) |
| #Coop. | 0.0(0.0) | 0.6(0.6) | 2.3(0.3) | 1.6(0.1) | 2.0(0.2) | 0.0(0.0) |
| #Hunt | 0.7(0.1) | 0.6(0.3) | 0.3(0.0) | 0.7(0.0) | 0.9(0.1) | 0.7(0.0) |
| PBT | w=[0.5,1] | w=[0,1] | w=[0,0] | RPG | RND | |
| Rew. | 3.3(0.2) | 4.8(0.6) | 5.1(0.4) | 6.0(0.5) | 8.9(0.3) | 3.2(0.2) |
| #Attack | 0.4(0.0) | 0.7(0.2) | 0.3(0.1) | 0.5(0.1) | 0.9(0.1) | 0.4(0.0) |
| #Coop. | 0.0(0.0) | 0.6(0.6) | 2.3(0.3) | 1.6(0.1) | 2.0(0.2) | 0.0(0.0) |
| #Hunt | 0.7(0.1) | 0.6(0.3) | 0.3(0.0) | 0.7(0.0) | 0.9(0.1) | 0.7(0.0) |
| Agent | Agent | Adapt. | apt. | |||
| Opponent:Cooperative → Competitive | ||||||
| #Attack | 0.2(0.0) | 0.3(0.0) | 0.1(0.1) | |||
| Rew. | 0.7(0.7) | -0.2(0.6) | 0.8(0.5) | |||
| Opponent: Competitive→Cooperative | ||||||
| #Coop. | 1.0(0.3) | 1.4(0.4) | 0.3(0.4) | |||
| Rew. | 2.5(0.7) | 3.6(1.2) | 1.1(0.7) | |||
| Agent | Agent | Adapt. | apt. | |||
| Opponent:Cooperative → Competitive | ||||||
| #Attack | 0.2(0.0) | 0.3(0.0) | 0.1(0.1) | |||
| Rew. | 0.7(0.7) | -0.2(0.6) | 0.8(0.5) | |||
| Opponent: Competitive→Cooperative | ||||||
| #Coop. | 1.0(0.3) | 1.4(0.4) | 0.3(0.4) | |||
| Rew. | 2.5(0.7) | 3.6(1.2) | 1.1(0.7) | |||
| Hyper-parameters | Value |
| Initial learning rate | 1e-3 |
| Minibatch size | 320 chunks of 10 timesteps |
| Adam stepsize (ε) | 1e-5 |
| Discount rate (γ) | 0.99 |
| GAE parameter (入) | 0.95 |
| Value loss coefficient | 1 |
| Entropy coefficient | 0.01 |
| Gradient clipping | 0.5 |
| PPO clipping parameter | 0.2 |
| Parallel threads | 64(Escalation),256(Monster-Hunt) |
| PPO epochs | 4 |
| reward scale parameter | 0.1 |
| episode length | 50 |
| Hyper-parameters | Value |
| Learning rate | 2.5e-4 |
| Minibatch size | 2 * 512 chunks of 32 timesteps |
| Adam stepsize (ε) | 1e-5 |
| Discount rate () | 0.995 |
| GAE parameter (入) | 0.95 |
| Value loss coefficient | 0.5 |
| action loss coefficient | 1 |
| Entropy coefficient | 0.01(discrete), 0.0025(continuous) |
| Gradient clipping | 20 |
| PPO clipping parameter | 0.1 |
| Parallel threads | 128 |
| PPO epochs | 4 |
| episode length | 128 |
| Hyper-parameters | Value |
| Initial learning rate | 1e-3 |
| Minibatch size | 320 chunks of 10 timesteps |
| Adam stepsize (ε) | 1e-5 |
| Discount rate (γ) | 0.99 |
| GAE parameter (入) | 0.95 |
| Value loss coefficient | 1 |
| Entropy coefficient | 0.01 |
| Gradient clipping | 0.5 |
| PPO clipping parameter | 0.2 |
| Parallel threads | 64(Escalation),256(Monster-Hunt) |
| PPO epochs | 4 |
| reward scale parameter | 0.1 |
| episode length | 50 |
| Hyper-parameters | Value |
| Learning rate | 2.5e-4 |
| Minibatch size | 2 * 512 chunks of 32 timesteps |
| Adam stepsize (ε) | 1e-5 |
| Discount rate () | 0.995 |
| GAE parameter (入) | 0.95 |
| Value loss coefficient | 0.5 |
| action loss coefficient | 1 |
| Entropy coefficient | 0.01(discrete), 0.0025(continuous) |
| Gradient clipping | 20 |
| PPO clipping parameter | 0.1 |
| Parallel threads | 128 |
| PPO epochs | 4 |
| episode length | 128 |
| Settings | Policy | Rewards | #Split | #Hunt | #Attack | #Cooperate |
| Standard | w=[1,1] RPG | 3.843(0.23) 4.34(0.171) | 0.859(0.083) 0.971(0.13) | 0.411(0.034) 0.659(0.048) | 0.526(0.064) 0.548(0.038) | 2.203(0.136) 2.028(0.297) |
| w=[0.5,1] | 3.827(0.489) | 0.807(0.192) | 0.365(0.106) | 0.15(0.064) | 2.342(0.286) | |
| w=[1,0.5] | 3.174(0.653) | 0.718(0.148) | 0.432(0.026) | 0.458(0.031) | 1.716(0.418) | |
| Original | 1.08(0.836) | 0.3(0.19) | 0.361(0.134) | 0.291(0.098) | 0.483(0.442) | |
| RND PBT | 2.789(0.346) | 0.499(0.061) | 0.623(0.128) | 0.242(0.037) | 1.349(0.164) | |
| Aggressive | 3.822(0.347) | 0.744(0.129) | 0.585(0.146) | 0.297(0.055) | 1.935(0.167) | |
| w=[0,0] | 5.966(0.539) | 1.195(0.155) | 0.699(0.008) | 0.517(0.066) | 1.603(0.127) | |
| RPG | 8.907(0.292) | 1.655(0.138) | 0.862(0.053) | 0.903(0.081) | 2.039(0.209) | |
| w=[0,1] | 5.066(0.375) | 0.785(0.041) | 0.344(0.049) | 0.346(0.058) | 2.327(0.311) | |
| w=[1,1] | 4.622(0.277) | 0.836(0.304) | 0.934(0.108) | 0.552(0.019) | 0.028(0.023) | |
| w=[0.5,1] | 4.79(0.588) | 0.678(0.31) | 0.617(0.28) | 0.67(0.194) | 0.55(0.643) | |
| Original | 3.551(0.121) | 0.717(0.032) | 0.812(0.078) | 0.412(0.018) | 0.027(0.026) | |
| RND | 3.189(0.154) | 0.626(0.065) | 0.705(0.008) | 0.382(0.029) | 0.035(0.027) | |
| PBT | 3.348(0.222) | 0.697(0.133) | 0.732(0.096) | 0.396(0.014) | 0.007(0.005) |
| Original | 1w=[4,0,0,0]w=[0,0,0,4]w=[0,4,4,0]w=[4,1,4,0] | ||||
| #Rewards | 20.00(0.00) | 74.76(2.88) | 20.00(0.00) | -470.0(0.00) | -453.45(0.25) |
| Settings | Policy | Rewards | #Split | #Hunt | #Attack | #Cooperate |
| Standard | w=[1,1] RPG | 3.843(0.23) 4.34(0.171) | 0.859(0.083) 0.971(0.13) | 0.411(0.034) 0.659(0.048) | 0.526(0.064) 0.548(0.038) | 2.203(0.136) 2.028(0.297) |
| w=[0.5,1] | 3.827(0.489) | 0.807(0.192) | 0.365(0.106) | 0.15(0.064) | 2.342(0.286) | |
| w=[1,0.5] | 3.174(0.653) | 0.718(0.148) | 0.432(0.026) | 0.458(0.031) | 1.716(0.418) | |
| Original | 1.08(0.836) | 0.3(0.19) | 0.361(0.134) | 0.291(0.098) | 0.483(0.442) | |
| RND PBT | 2.789(0.346) | 0.499(0.061) | 0.623(0.128) | 0.242(0.037) | 1.349(0.164) | |
| Aggressive | 3.822(0.347) | 0.744(0.129) | 0.585(0.146) | 0.297(0.055) | 1.935(0.167) | |
| w=[0,0] | 5.966(0.539) | 1.195(0.155) | 0.699(0.008) | 0.517(0.066) | 1.603(0.127) | |
| RPG | 8.907(0.292) | 1.655(0.138) | 0.862(0.053) | 0.903(0.081) | 2.039(0.209) | |
| w=[0,1] | 5.066(0.375) | 0.785(0.041) | 0.344(0.049) | 0.346(0.058) | 2.327(0.311) | |
| w=[1,1] | 4.622(0.277) | 0.836(0.304) | 0.934(0.108) | 0.552(0.019) | 0.028(0.023) | |
| w=[0.5,1] | 4.79(0.588) | 0.678(0.31) | 0.617(0.28) | 0.67(0.194) | 0.55(0.643) | |
| Original | 3.551(0.121) | 0.717(0.032) | 0.812(0.078) | 0.412(0.018) | 0.027(0.026) | |
| RND | 3.189(0.154) | 0.626(0.065) | 0.705(0.008) | 0.382(0.029) | 0.035(0.027) | |
| PBT | 3.348(0.222) | 0.697(0.133) | 0.732(0.096) | 0.396(0.014) | 0.007(0.005) |
| Original | 1w=[4,0,0,0]w=[0,0,0,4]w=[0,4,4,0]w=[4,1,4,0] | ||||
| #Rewards | 20.00(0.00) | 74.76(2.88) | 20.00(0.00) | -470.0(0.00) | -453.45(0.25) |
| Oppo. Type | Stag | Hare | TFT | Random |
| #Stag | 9.31(0.77) | 3.6(4.33) | 7.31(3.82) | 5.35(3.48) |
| #Hare | 0.69(0.77) | 6.4(4.33) | 2.69(3.81) | 4.65(3.48) |
| Oppo. Type | Stag | Hare | TFT | Random |
| #Stag | 9.31(0.77) | 3.6(4.33) | 7.31(3.82) | 5.35(3.48) |
| #Hare | 0.69(0.77) | 6.4(4.33) | 2.69(3.81) | 4.65(3.48) |
| Stag | Hare | |
| Stag | a,a | c,b |
| Hare | b,c | d,d |
| PBT | RR | RPG | RND | |
| Rew. | 3.8(0.3) | 3.8(0.2) | 4.3(0.2) | 2.8(0.3) |
| #Coop. | 1.9(0.2) | 2.2(0.1) | 2.0(0.3) | 1.3(0.2) |
| #Hunt | 0.6(0.1) | 0.4(0.0) | 0.7(0.0) | 0.6(0.1) |
| Chase Agent1 Respawn Move together Move together | |||||||||||||||||||
| 回 | 一 | ||||||||||||||||||
| 一 | 8 | 8 | 8 | 8 | |||||||||||||||
| 一 | 一 | 一 | 一 | ? | |||||||||||||||
| 六 | 电 | ℃ | |||||||||||||||||
| Two agents step on the same grid. Respawn Chase Agent | |||||||||||||||||||
| PBT | w=[0.5,1] | w=[0,1] | w=[0,0] | RPG | RND | |
| Rew. | 3.3(0.2) | 4.8(0.6) | 5.1(0.4) | 6.0(0.5) | 8.9(0.3) | 3.2(0.2) |
| #Attack | 0.4(0.0) | 0.7(0.2) | 0.3(0.1) | 0.5(0.1) | 0.9(0.1) | 0.4(0.0) |
| #Coop. | 0.0(0.0) | 0.6(0.6) | 2.3(0.3) | 1.6(0.1) | 2.0(0.2) | 0.0(0.0) |
| #Hunt | 0.7(0.1) | 0.6(0.3) | 0.3(0.0) | 0.7(0.0) | 0.9(0.1) | 0.7(0.0) |
| Oppo. | M. | M-Coop. | M-Alone. | Apple. |
| #C-H | 16.3(19.2) | 20.9(0.8) | 14.2(18.0) | 2.7(1.0) |
| #S-H | 1.2(0.4) | 0.4(0.1) | 2.2(1.2) | 2.2(1.4) |
| #Apple | 12.4(7.3) | 3.3(0.8) | 10.9(7.0) | 13.6(3.8) |
| Agent | Agent | Adapt. | apt. | |||
| Opponent:Cooperative → Competitive | ||||||
| #Attack | 0.2(0.0) | 0.3(0.0) | 0.1(0.1) | |||
| Rew. | 0.7(0.7) | -0.2(0.6) | 0.8(0.5) | |||
| Opponent: Competitive→Cooperative | ||||||
| #Coop. | 1.0(0.3) | 1.4(0.4) | 0.3(0.4) | |||
| Rew. | 2.5(0.7) | 3.6(1.2) | 1.1(0.7) | |||
| Hyper-parameters | Value |
| Initial learning rate | 1e-3 |
| Minibatch size | 320 chunks of 10 timesteps |
| Adam stepsize (ε) | 1e-5 |
| Discount rate (γ) | 0.99 |
| GAE parameter (入) | 0.95 |
| Value loss coefficient | 1 |
| Entropy coefficient | 0.01 |
| Gradient clipping | 0.5 |
| PPO clipping parameter | 0.2 |
| Parallel threads | 64(Escalation),256(Monster-Hunt) |
| PPO epochs | 4 |
| reward scale parameter | 0.1 |
| episode length | 50 |
| Hyper-parameters | Value |
| Learning rate | 2.5e-4 |
| Minibatch size | 2 * 512 chunks of 32 timesteps |
| Adam stepsize (ε) | 1e-5 |
| Discount rate () | 0.995 |
| GAE parameter (入) | 0.95 |
| Value loss coefficient | 0.5 |
| action loss coefficient | 1 |
| Entropy coefficient | 0.01(discrete), 0.0025(continuous) |
| Gradient clipping | 20 |
| PPO clipping parameter | 0.1 |
| Parallel threads | 128 |
| PPO epochs | 4 |
| episode length | 128 |
| Settings | Policy | Rewards | #Split | #Hunt | #Attack | #Cooperate |
| Standard | w=[1,1] RPG | 3.843(0.23) 4.34(0.171) | 0.859(0.083) 0.971(0.13) | 0.411(0.034) 0.659(0.048) | 0.526(0.064) 0.548(0.038) | 2.203(0.136) 2.028(0.297) |
| w=[0.5,1] | 3.827(0.489) | 0.807(0.192) | 0.365(0.106) | 0.15(0.064) | 2.342(0.286) | |
| w=[1,0.5] | 3.174(0.653) | 0.718(0.148) | 0.432(0.026) | 0.458(0.031) | 1.716(0.418) | |
| Original | 1.08(0.836) | 0.3(0.19) | 0.361(0.134) | 0.291(0.098) | 0.483(0.442) | |
| RND PBT | 2.789(0.346) | 0.499(0.061) | 0.623(0.128) | 0.242(0.037) | 1.349(0.164) | |
| Aggressive | 3.822(0.347) | 0.744(0.129) | 0.585(0.146) | 0.297(0.055) | 1.935(0.167) | |
| w=[0,0] | 5.966(0.539) | 1.195(0.155) | 0.699(0.008) | 0.517(0.066) | 1.603(0.127) | |
| RPG | 8.907(0.292) | 1.655(0.138) | 0.862(0.053) | 0.903(0.081) | 2.039(0.209) | |
| w=[0,1] | 5.066(0.375) | 0.785(0.041) | 0.344(0.049) | 0.346(0.058) | 2.327(0.311) | |
| w=[1,1] | 4.622(0.277) | 0.836(0.304) | 0.934(0.108) | 0.552(0.019) | 0.028(0.023) | |
| w=[0.5,1] | 4.79(0.588) | 0.678(0.31) | 0.617(0.28) | 0.67(0.194) | 0.55(0.643) | |
| Original | 3.551(0.121) | 0.717(0.032) | 0.812(0.078) | 0.412(0.018) | 0.027(0.026) | |
| RND | 3.189(0.154) | 0.626(0.065) | 0.705(0.008) | 0.382(0.029) | 0.035(0.027) | |
| PBT | 3.348(0.222) | 0.697(0.133) | 0.732(0.096) | 0.396(0.014) | 0.007(0.005) |
| Original | 1w=[4,0,0,0]w=[0,0,0,4]w=[0,4,4,0]w=[4,1,4,0] | ||||
| #Rewards | 20.00(0.00) | 74.76(2.88) | 20.00(0.00) | -470.0(0.00) | -453.45(0.25) |
| Oppo. Type | Stag | Hare | TFT | Random |
| #Stag | 9.31(0.77) | 3.6(4.33) | 7.31(3.82) | 5.35(3.48) |
| #Hare | 0.69(0.77) | 6.4(4.33) | 2.69(3.81) | 4.65(3.48) |
| Data Relation | MovieLens X1-Count | Flickr | ||||
| X1-Binary | X2-Binary | X3-Binary | X4-Binary | X5-Gaussian | ||
| CCMF (reg=10) | 1.0588 | 0.7071 | 0.2473 | 0.3661 | 0.2384 | 1.0033 |
| CMF without bias | 1.0569 | 0.5120 | 0.2324 | 0.5093 | 0.2176 | 1.0092 |
| CMF with bias | 0.9475 | 0.5000 | 0.2369 | 0.2789 | 0.2109 | 1.0033 |
| gCMF without bias | 1.0418 | 0.5003 | 0.2291 | 0.5014 | 0.2167 | 1.0039 |
| gCMF with bias | 0.9474 | 0.5000 | 0.2369 | 0.2789 | 0.2109 | 1.0033 |
| Data Relation | MovieLens X1-Count | Flickr | ||||
| X1-Binary | X2-Binary | X3-Binary | X4-Binary | X5-Gaussian | ||
| CCMF (reg=10) | 1.0588 | 0.7071 | 0.2473 | 0.3661 | 0.2384 | 1.0033 |
| CMF without bias | 1.0569 | 0.5120 | 0.2324 | 0.5093 | 0.2176 | 1.0092 |
| CMF with bias | 0.9475 | 0.5000 | 0.2369 | 0.2789 | 0.2109 | 1.0033 |
| gCMF without bias | 1.0418 | 0.5003 | 0.2291 | 0.5014 | 0.2167 | 1.0039 |
| gCMF with bias | 0.9474 | 0.5000 | 0.2369 | 0.2789 | 0.2109 | 1.0033 |
| Data Relation | MovieLens X1-Count | Flickr | ||||
| X1-Binary | X2-Binary | X3-Binary | X4-Binary | X5-Gaussian | ||
| CCMF (reg=10) | 1.0588 | 0.7071 | 0.2473 | 0.3661 | 0.2384 | 1.0033 |
| CMF without bias | 1.0569 | 0.5120 | 0.2324 | 0.5093 | 0.2176 | 1.0092 |
| CMF with bias | 0.9475 | 0.5000 | 0.2369 | 0.2789 | 0.2109 | 1.0033 |
| gCMF without bias | 1.0418 | 0.5003 | 0.2291 | 0.5014 | 0.2167 | 1.0039 |
| gCMF with bias | 0.9474 | 0.5000 | 0.2369 | 0.2789 | 0.2109 | 1.0033 |