facebookresearch/vjepa2

View on GitHub

PyTorch code and models for VJEPA2 self-supervised learning from video.

Stars: 4.6k
Forks: 573
Open beginner issues: 0
Indexed issues: 73

Dominant language: Python
License: MIT
Last GitHub push: Mar 23, 2026
Latest indexed: Sep 15, 2026
Contributing guide: Contributing guide
Code of conduct: Code of conduct
Beginner labels: No beginner labels indexed
PR merge metrics: No merged PRs in 30d

Open All

73 open issues indexed

Exploring Extensions of Your Work to Diverse Robot Platforms Open

facebookresearch/vjepa2#38 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
OSError: facebook/vjepa2-vitg-fpc64-384 does not appear to have a file named pytorch_model.bin, model.safetensors, tf_model.h5, model.ckpt or flax_model.msgpack. Open

facebookresearch/vjepa2#46 · 5 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Workflow for fine-tuning from a pretrained checkpoint Open

facebookresearch/vjepa2#47 · 4 comments · 1 reaction · 0 assignees · Indexed Sep 15, 2026
Pretraining Error:Unexpected key(s) in state_dict: "module.backbone.mask_tokens.6" Open

facebookresearch/vjepa2#49 · 4 comments · 1 reaction · 0 assignees · Indexed Sep 15, 2026
Question about the prediction ability about V-JEPA 2-AC checkpoints in energy_landscape_example.ipynb Open

facebookresearch/vjepa2#54 · 7 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Validating VJEPA2-AC model Open

facebookresearch/vjepa2#57 · 0 comments · 3 reactions · 0 assignees · Indexed Sep 15, 2026
Discrepancy Between Paper and Configuration in V-JEPA 2-AC Model Open

facebookresearch/vjepa2#64 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
A question about vjepa2-ac when I want to get new action. Open

facebookresearch/vjepa2#66 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
inference on variable number of frames Open

facebookresearch/vjepa2#68 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
inference with encoder model consumes too large memory Open

facebookresearch/vjepa2#71 · 5 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Training loss suddenly drops after a while of training Open

facebookresearch/vjepa2#74 · 6 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Acc=0 occurs during training for evals Open

facebookresearch/vjepa2#76 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
About planning in real world tasks Open

facebookresearch/vjepa2#79 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Project hardcodes "cuda", not allowing the user to specify another device type Open

facebookresearch/vjepa2#80 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
finetune vjepa2-ac on my dataset Open

facebookresearch/vjepa2#84 · 5 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Question about encoder inputs of the vjepa2-ac model Open

facebookresearch/vjepa2#87 · 1 comment · 1 reaction · 0 assignees · Indexed Sep 15, 2026
Training decoder to visualize vjepa representations Open

facebookresearch/vjepa2#88 · 9 comments · 2 reactions · 0 assignees · Indexed Sep 15, 2026
Train log Open

facebookresearch/vjepa2#89 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Patching and Predictor Size Open

facebookresearch/vjepa2#90 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
LLM alignment for Question-Answering Open

facebookresearch/vjepa2#92 · 2 comments · 15 reactions · 0 assignees · Indexed Sep 15, 2026
Guidance on fine-tuning V-JEPA 2 Open

facebookresearch/vjepa2#93 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Convert trained V-JEPA 2 model to huggingface model Open

facebookresearch/vjepa2#96 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Does vjepa2 support videos with arbitrary aspect ratios (such as 1:2 or 2:1)? Will the performance be affected? Open

facebookresearch/vjepa2#97 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Eval of Jester Dataset Open

facebookresearch/vjepa2#99 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Flexibility in Frame Sampling Open

facebookresearch/vjepa2#102 · 0 comments · 2 reactions · 0 assignees · Indexed Sep 15, 2026
The regression error of predictor is very high Open

facebookresearch/vjepa2#104 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Processing image data for V-JEPA2 for QA Open

facebookresearch/vjepa2#105 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Bug in clip resizing Open

facebookresearch/vjepa2#106 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Stereo videos in DROID raw: included in V-JEPA 2-AC training? Open

facebookresearch/vjepa2#107 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Transforms for the QA Dataset Open

facebookresearch/vjepa2#108 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
AC model not working in simulation Open

facebookresearch/vjepa2#110 · 7 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
如何将我的后训练vjepa2部署到我自己的机械臂（ac版本的模型也是我基于我的数据集构造的） Open

facebookresearch/vjepa2#111 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Add Lightweight Backbones to Enable Applied & Low‑Resource Research Open

facebookresearch/vjepa2#112 · 0 comments · 2 reactions · 0 assignees · Indexed Sep 15, 2026
Temporal features from frames Open

facebookresearch/vjepa2#117 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
tubelet_size and frameskip Open

facebookresearch/vjepa2#118 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Using VJEPA2 Frozen Encoder as Video Feature Extractor Open

facebookresearch/vjepa2#120 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
How can I use multi-layer feature maps from the V-JEPA model to perform classification tasks such as ImageNet? Does the repo provide any example code for this? I couldn’t find it Open

facebookresearch/vjepa2#122 · 0 comments · 1 reaction · 0 assignees · Indexed Sep 15, 2026
Predictor mask token mismatch Open

facebookresearch/vjepa2#124 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Typo in the chart? Open

facebookresearch/vjepa2#131 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Mismatch in Teacher Forcing Steps (k) Between Paper and Code for the vjepa2-ac Predictor Head Open

facebookresearch/vjepa2#132 · 0 comments · 3 reactions · 0 assignees · Indexed Sep 15, 2026
weights 2.1? Open

facebookresearch/vjepa2#134 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Typo in chart legend: V-JEPA 2.5 should be 2.1 Open

facebookresearch/vjepa2#135 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Training V-JEPA 2 on images Open

facebookresearch/vjepa2#136 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Release V-JEPA 2.1 on Hugging Face Open

facebookresearch/vjepa2#137 · 5 comments · 9 reactions · 0 assignees · Indexed Sep 15, 2026
Torch Hub broken due to localhost URL in repo Open

facebookresearch/vjepa2#138 · 1 comment · 5 reactions · 0 assignees · Indexed Sep 15, 2026
Training loss gets into Equilibrium early, then rises while downstream accuracy also degrades on Sentinel-2 temporal pretraining Open

facebookresearch/vjepa2#140 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Clarification needed: image results show V-JEPA 2.5 instead of V-JEPA 2.1 Open

facebookresearch/vjepa2#141 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Duplicate encoders loaded in vjepa2-ac training Open

facebookresearch/vjepa2#144 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
codes for downstream tasks Open

facebookresearch/vjepa2#145 · 2 comments · 4 reactions · 0 assignees · Indexed Sep 15, 2026
Dense Features Extractions with V-JEPA 2.1 Open

facebookresearch/vjepa2#146 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026

facebookresearch/vjepa2

Get new issues in your inbox