Multimodal Genre Classification on MM-IMDb 1.0 (test)
65.34Macro F1Ours (GW, BLIP-2)
Evaluation Results
| Method | Links | |
|---|---|---|
| Ours (GW, BLIP-2)End-to-end=no, Backbone=BLIP-2, Train FLOPs (est.)=7.5 * 10^142026.02 | 65.34 | |
| BridgeTow2026.02 | 63.3 | |
| CoMM (BLIP-2, finetuned)End-to-end=yes, Backbone=BLIP-2, Mode=finetuned, Train FLOPs (est.)=> 4.7 * 10^152026.02 | 62 | |
| Ours (GW, BLIP-2, random)End-to-end=no, Backbone=BLIP-2, Strategy=random, Train FLOPs (est.)=7.5 * 10^142026.02 | 60.41 | |
| CoMM (CLIP, finetuned)End-to-end=yes, Backbone=CLIP, Mode=finetuned2026.02 | 58.97 | |
| CoMM (BLIP-2, linear readout)End-to-end=no, Backbone=BLIP-2, Mode=linear readout, Train FLOPs (est.)=4.7 * 10^152026.02 | 58.44 | |
| MFASEnd-to-end=yes2026.02 | 55.6 | |
| CoMM (CLIP, linear readout)End-to-end=no, Backbone=CLIP, Mode=linear readout2026.02 | 54.63 | |
| GMUEnd-to-end=yes, Backbone=Arevalo et al., 20172026.02 | 54.1 | |
| Ours (GW, CLIP)End-to-end=no, Backbone=CLIP2026.02 | 53.48 | |
| CLIPEnd-to-end=no, Backbone=CLIP2026.02 | 50.9 | |
| Ours (GW, CLIP, random)End-to-end=no, Backbone=CLIP, Strategy=random2026.02 | 50 | |
| BLIP-2End-to-end=no, Backbone=BLIP-22026.02 | 49.9 |