visbench-semantic_segmentation-dinov2_vits14

semantic_segmentation probe for dinov2_vits14

A trained probe head, not a backbone. It is the small module VisBench fits on top of frozen dinov2_vits14 features to measure what those features carry.

import visbench
from visbench.hub import load_probe_from_hub

backbone = visbench.get_backbone("dinov2_vits14")
probe = load_probe_from_hub("turhancan97/visbench-semantic_segmentation-dinov2_vits14", backbone=backbone)

It only works with this backbone

These weights were fitted on features from dinov2_vits14, taken with pooling=mean and feature_mode=dense_only. Loading them against anything else is refused, because the failure is otherwise silent: a head fitted on one pooling and fed another has the right shapes and produces a plausible, wrong number.

Reported scores

Reproducing it

Fitted with:

• batch_size: 8
• epochs: 10
• head: linear
• hidden_dim: 512
• ignore_index: -1
• layers: None
• loss: cross_entropy
• lr: 0.0005
• miou_reduction: dataset_and_per_image
• num_classes: 21
• optimizer: adamw
• protocol: visbench_semantic_seg
• warmup_epochs: 1.5
• weight_decay: 0.0001

Generated by VisBench.