Search results

Replicating geobench results

Status: Open.

#310 In facebookresearch/dinov3;

· raviolihui opened on Jan 29, 2026

pretrain on imagenet，The training results are inconsistent with the repo results

Status: Open.

#309 In facebookresearch/dinov3;

· Elijah-Yi opened on Jan 27, 2026

Lightweight Semantic Segmentation

Status: Open.

#308 In facebookresearch/dinov3;

· ovunctuzel-bc opened on Jan 26, 2026

Clarification of the zero-shot segmentation baselines in Table 16

Status: Open.

#307 In facebookresearch/dinov3;

· benji2264 opened on Jan 26, 2026

Can I continue pre-train DINOv3 using the script dinov3/train/train.py?

Status: Open.

#306 In facebookresearch/dinov3;

· yangguoquan001 opened on Jan 21, 2026

how to get high-resolution feature?

Status: Open.

#305 In facebookresearch/dinov3;

· YWliu040120 opened on Jan 19, 2026

Configuration for knowledge distillation from DINOv3 7B

Status: Open.

#304 In facebookresearch/dinov3;

· NicoSchulthess opened on Jan 15, 2026

Could you please release the code used to train the M2F segmentation head?

Status: Open.

#300 In facebookresearch/dinov3;

· argmaxlife opened on Jan 4, 2026

Shape mismatch when loading the pre-trained weights in Stage1

Status: Open.

#299 In facebookresearch/dinov3;

· moothes opened on Jan 2, 2026

Step-wise loss descent correlates with do_test frequency; removing do_test leads to loss stagnation

Status: Open.

#298 In facebookresearch/dinov3;

· gaofan3 opened on Jan 2, 2026

If I only have one GPU, could I increase something like accumulation_steps for obtaining larger batch size?

Status: Open.

#296 In facebookresearch/dinov3;

· vicdxxx opened on Dec 23, 2025

why num_feature_levels is 1 in _make_dinov3_detector in dinov3/hub/detectors.py, sould it be 4?

Status: Open.

#295 In facebookresearch/dinov3;

· YifengDeng1 opened on Dec 18, 2025