# DistInit: Learning Video Representations Without a Single Labeled Video

arXiv: https://arxiv.org/abs/1901.09244 · HTML: https://mlsanity.com/paper/1901.09244/

## Key facts

- Cited by 21 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 26 Jan 2019.

## Built on by (most cited first, 21 of 21)

- [NeSF: Neural Semantic Fields for Generalizable Semantic Segmentation of 3D Scenes](https://mlsanity.com/paper/2111.13260/) (2021)
- [Ego-Exo: Transferring Visual Representations from Third-person to First-person Videos](https://mlsanity.com/paper/2104.07905/) (2021)
- [EgoExoLearn: A Dataset for Bridging Asynchronous Ego- and Exo-centric View of Procedural Activities in Real World](https://mlsanity.com/paper/2403.16182/) (2024)
- [Learning Audio-Video Modalities from Image Captions](https://mlsanity.com/paper/2204.00679/) (2022)
- [Learning 3D Semantic Segmentation with only 2D Image Supervision](https://mlsanity.com/paper/2110.11325/) (2021)
- [Distilling Audio-Visual Knowledge by Compositional Contrastive Learning](https://mlsanity.com/paper/2104.10955/) (2021)
- [Learning from 2D: Contrastive Pixel-to-Point Knowledge Transfer for 3D Pretraining](https://mlsanity.com/paper/2104.04687/) (2021)
- [Multiview Pseudo-Labeling for Semi-supervised Learning from Video](https://mlsanity.com/paper/2104.00682/) (2021)
- [Motion-Augmented Self-Training for Video Recognition at Smaller Scale](https://mlsanity.com/paper/2105.01646/) (2021)
- [FitCLIP: Refining Large-Scale Pretrained Image-Text Models for Zero-Shot Video Understanding Tasks](https://mlsanity.com/paper/2203.13371/) (2022)
- [TimeBalance: Temporally-Invariant and Temporally-Distinctive Video Representations for Semi-Supervised Action Recognition](https://mlsanity.com/paper/2303.16268/) (2023)
- [NIL: No-data Imitation Learning by Leveraging Pre-trained Video Diffusion Models](https://mlsanity.com/paper/2503.10626/) (2025)
- [Network Pruning via Feature Shift Minimization](https://mlsanity.com/paper/2207.02632/) (2022)
- [Learning Discriminative Spatio-temporal Representations for Semi-supervised Action Recognition](https://mlsanity.com/paper/2404.16416/) (2024)
- [Distillation of Human-Object Interaction Contexts for Action Recognition](https://mlsanity.com/paper/2112.09448/) (2021)
- [Seeing your sleep stage: cross-modal distillation from EEG to infrared video](https://mlsanity.com/paper/2208.05814/) (2022)
- [Network Compression via Central Filter](https://mlsanity.com/paper/2112.05493/) (2021)
- [iBoot: Image-bootstrapped Self-Supervised Video Representation Learning](https://mlsanity.com/paper/2206.08339/) (2022)
- [SoccerKDNet: A Knowledge Distillation Framework for Action Recognition in Soccer Videos](https://mlsanity.com/paper/2307.07768/) (2023)
- [Seeing in the Dark: A Teacher-Student Framework for Dark Video Action Recognition via Knowledge Distillation and Contrastive Learning](https://mlsanity.com/paper/2502.03724/) (2025)
- [LEViL: Label-Efficient Video Learning via Zero-Shot Distillation over VLM-Generated Pseudo-Label Spaces](https://mlsanity.com/paper/2606.21358/) (2026)

## Abstract (arXiv)

Video recognition models have progressed significantly over the past few years, evolving from shallow classifiers trained on hand-crafted features to deep spatiotemporal networks. However, labeled video data required to train such models have not been able to keep up with the ever-increasing depth and sophistication of these networks. In this work, we propose an alternative approach to learning video representations that require no semantically labeled videos and instead leverages the years of effort in collecting and labeling large and clean still-image datasets. We do so by using state-of-the-art models pre-trained on image datasets as "teachers" to train video models in a distillation framework. We demonstrate that our method learns truly spatiotemporal features, despite being trained only using supervision from still-image networks. Moreover, it learns good representations across different input modalities, using completely uncurated raw video data sources and with different 2D teacher models. Our method obtains strong transfer performance, outperforming standard techniques for bootstrapping video architectures with image-based models by 16%. We believe that our approach opens up new approaches for learning spatiotemporal representations from unlabeled video data.