Multi-object semantic discovery on CLEVR
70.6mAccOurs-D
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Ours-DSprite generation=MLP, Decision function=linear mapping, Activation function=GS, Training criterion=Lcomp, Regularization=Lempty2026.04 | 70.6 | 55.3 | |
| Ours-DSprite generation=MLP, Decision function=linear mapping, Activation function=GS, Training criterion=Lcomp, Regularization=Lempty + Lfreq2026.04 | 70.5 | 53.9 | |
| DTI-SpritesSprite generation=Pixels, Decision function=Min-Loss, Activation function=S, Training criterion=L0−1=comp, Regularization=Lempty, Training schedule=Original paper training schedule (italic)2026.04 | 69.8 | 55.7 | |
| Ours-DSprite generation=MLP, Decision function=linear mapping, Activation function=GS, Training criterion=Lcomp, Regularization=Lempty + Lbin2026.04 | 68 | 53.1 | |
| Ours-DSprite generation=MLP, Decision function=linear mapping, Activation function=GS, Training criterion=Lcomp, Regularization=None2026.04 | 59.3 | 43 | |
| Ours-DSprite generation=MLP, Decision function=linear mapping, Activation function=GS, Training criterion=Lcomp, Regularization=τ2026.04 | 55.3 | 42.1 |