Referring Expression Comprehension on RefGTA (test)
72.56AccuracyReCLIP w/o relations
Evaluation Results
| Method | Links | |
|---|---|---|
| ReCLIP w/o relationsBackbone=Ensemble of RN50x16 and ViT-B/32, Proposals=Ground-truth (GT)2022.04 | 72.56 | |
| ReCLIP w/o relationsBackbone=Ensemble of RN50x16 and ViT-B/32, Proposals=Detected proposals filtered to only people (DT-P)2022.04 | 70.84 | |
| ReCLIPBackbone=Ensemble of RN50x16 and ViT-B/32, Proposals=Ground-truth (GT)2022.04 | 70.79 | |
| ReCLIPBackbone=Ensemble of RN50x16 and ViT-B/32, Proposals=Detected proposals filtered to only people (DT-P)2022.04 | 69.05 | |
| ReCLIPBackbone=Ensemble of RN50x16 and ViT-B/32, Proposals=Detected proposals (DT)2022.04 | 61.38 | |
| ReCLIP w/o relationsBackbone=Ensemble of RN50x16 and ViT-B/32, Proposals=Detected proposals (DT)2022.04 | 61.31 | |
| MDETRPretrained status=Not finetuned, Backbone=EfficientNet-B3, Proposals=None (End-to-End)2022.04 | 56.6 | |
| CLIP GradCAMBackbone=Ensemble of RN50x16 and ViT-B/32, Proposals=Ground-truth (GT)2022.04 | 51.53 | |
| UNITER-LargeFinetuning Dataset=RefCOCOg, Proposals=Ground-truth (GT)2022.04 | 51.05 | |
| CLIP GradCAMBackbone=Ensemble of RN50x16 and ViT-B/32, Proposals=Detected proposals filtered to only people (DT-P)2022.04 | 50.73 | |
| UNITER-LargeFinetuning Dataset=RefCOCO+, Proposals=Ground-truth (GT)2022.04 | 50.6 | |
| UNITER-LargeFinetuning Dataset=RefCOCOg, Proposals=Detected proposals filtered to only people (DT-P)2022.04 | 49.78 | |
| UNITER-LargeFinetuning Dataset=RefCOCO+, Proposals=Detected proposals filtered to only people (DT-P)2022.04 | 48.3 | |
| MDETRFinetuning Dataset=RefCOCOg, Backbone=EfficientNet-B3, Proposals=None (End-to-End)2022.04 | 39.13 | |
| MDETRFinetuning Dataset=RefCOCO+, Backbone=EfficientNet-B3, Proposals=None (End-to-End)2022.04 | 39.02 | |
| CLIP GradCAMBackbone=Ensemble of RN50x16 and ViT-B/32, Proposals=Detected proposals (DT)2022.04 | 34.51 | |
| UNITER-LargeFinetuning Dataset=RefCOCO+, Proposals=Detected proposals (DT)2022.04 | 34.4 | |
| UNITER-LargeFinetuning Dataset=RefCOCOg, Proposals=Detected proposals (DT)2022.04 | 28.31 | |
| RandomProposals=Ground-truth (GT)2022.04 | 27.6 | |
| RandomProposals=Detected proposals filtered to only people (DT-P)2022.04 | 21.75 | |
| RandomProposals=Detected proposals (DT)2022.04 | 5.13 |