Multi-Task Learning on PASCAL-Context (val)
72.9SemSeg mIoUTGLoRA
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TGLoRABackbone=PVT-Small, Method configuration=ATTN, Trainable Params (M)=3.852025.09 | 72.9 | 60.15 | 66.76 | 16.91 | 3.57 | |
| HyperFormerBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=72.772026.03 | 71.43 | 60.73 | 65.54 | 17.77 | 2.64 | |
| Free SinewichBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Rank (r)=64, Trainable Parameters (M)=6.532026.03 | 71.25 | 61.38 | 66.24 | 16.14 | 5.39 | |
| Free SinewichBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Rank (r)=32, Trainable Parameters (M)=4.042026.03 | 71.02 | 60.75 | 65.94 | 16.44 | 4.51 | |
| PolyhistorBackbone=PVT-Small, Trainable Params (M)=7.322025.09 | 71 | 57.52 | 65.83 | 17.83 | 0.13 | |
| Free SinewichBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Rank (r)=16, Trainable Parameters (M)=2.792026.03 | 70.92 | 59.78 | 65.32 | 16.7 | 3.47 | |
| PolyhistorBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=8.962026.03 | 70.87 | 59.54 | 65.47 | 17.47 | 2.34 | |
| TADFormerBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Rank (r)=64, Trainable Parameters (M)=7.382026.03 | 70.82 | 60.45 | 65.88 | 16.48 | 4.24 | |
| HyperformerBackbone=PVT-Small, Trainable Params (M)=16.142025.09 | 70.81 | 57.76 | 65.49 | 17.75 | 0.14 | |
| MTLoRA (Reproduced)Backbone=PVT-Small, Method configuration=ATTN, Trainable Params (M)=4.402025.09 | 70.34 | 58.83 | 65.89 | 17.15 | 1.42 | |
| MTLoRA (Reproduced)Backbone=PVT-Small, Method configuration=ATTN + SR, Trainable Params (M)=8.692025.09 | 70.32 | 59.06 | 66.1 | 16.99 | 1.81 | |
| VL-AdapterBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=4.742026.03 | 70.21 | 59.15 | 62.29 | 19.26 | -1.83 | |
| TADFormerBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Rank (r)=32, Trainable Parameters (M)=4.782026.03 | 70.2 | 60 | 65.71 | 16.57 | 3.63 | |
| LoRABackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=2.872026.03 | 70.12 | 57.73 | 61.9 | 18.96 | -2.17 | |
| DiTASK - MTLBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=3.552026.03 | 70.09 | 59.03 | 64.55 | 17.47 | 1.47 | |
| TADFormerBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Rank (r)=16, Trainable Parameters (M)=3.562026.03 | 69.79 | 59.27 | 65.04 | 16.91 | 2.44 | |
| MTLoRABackbone=PVT-Small, r=64, Trainable Params (M)=8.692025.09 | 69.74 | 58.08 | 65.62 | 17.35 | 1.2 | |
| DiTASK - MTL*Backbone=Swin Transformer Tiny, Pre-trained=ImageNet-22k, Trainable Parameters (M)=3.552026.03 | 69.66 | 62.02 | 65 | 17.1 | 3.22 | |
| AdapterBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=11.242026.03 | 69.21 | 57.38 | 61.28 | 18.83 | -2.71 | |
| Single Task – Full Fine-TuningBackbone=PVT-Small, Trainable Params (M)=97.512025.09 | 68.81 | 61.27 | 62.67 | 17.55 | 0 | |
| BitfitBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=2.852026.03 | 68.57 | 55.99 | 60.64 | 19.42 | -4.6 | |
| MTLoRABackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Rank (r)=16, Trainable Parameters (M)=4.952026.03 | 68.19 | 58.99 | 64.48 | 17.03 | 1.35 | |
| CompactorBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=2.782026.03 | 68.08 | 56.41 | 60.08 | 19.22 | -4.55 | |
| MTLoRABackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Rank (r)=64, Trainable Parameters (M)=8.342026.03 | 67.9 | 59.84 | 65.4 | 16.6 | 2.55 | |
| MTLoRABackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Rank (r)=32, Trainable Parameters (M)=6.082026.03 | 67.74 | 59.46 | 64.9 | 16.59 | 2.16 | |
| MTL - Full Fine TuningBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=30.062026.03 | 67.56 | 60.24 | 65.21 | 16.64 | 2.23 | |
| Compactor++Backbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=2.662026.03 | 67.26 | 55.69 | 59.47 | 19.54 | -5.84 | |
| Single TaskBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=112.622026.03 | 67.21 | 61.93 | 62.35 | 17.97 | 0 | |
| MTL - Tuning Decoders OnlyBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=1.942026.03 | 65.09 | 53.48 | 57.46 | 20.69 | -9.95 | |
| MTL – Decoders OnlyBackbone=PVT-Small, Trainable Params (M)=2.112025.09 | 64.86 | 51.18 | 61.54 | 19.55 | 8.85 | |
| VPT-deepBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=3.432026.03 | 64.35 | 52.54 | 58.15 | 21.07 | -10.85 | |
| VPT-shallowBackbone=Swin Transformer Tiny, Pre-trained=ImageNet-1K, Trainable Parameters (M)=2.572026.03 | 62.96 | 52.27 | 58.31 | 20.9 | -11.18 |