Add dataset card

README.md

@@ -0,0 +1,78 @@
+---
+license: apache-2.0
+task_categories:
+  - video-classification
+  - feature-extraction
+tags:
+  - video
+  - tubelets
+  - vjepa
+  - egocentric
+  - raw-frames
+  - computer-vision
+size_categories:
+  - 10K<n<100K
+---
+
+# VJEPA Raw Tubelet Dataset (L2-Norm / CTD)
+
+Pre-extracted raw video tubelets from the [Egocentric-10K](https://huggingface.co/datasets/builddotai/Egocentric-10K) dataset, processed using the canonical **V-JEPA preprocessing pipeline**.
+
+## Tensor Format
+
+Each  file is a single tubelet saved with  and can be loaded with .
+
+| Property | Value |
+|----------|-------|
+| **Shape** |  |
+| **Dtype** |  |
+| **Layout** |  — Frames × Height × Width × Channels |
+| **Frames** | 64 raw RGB frames per tubelet |
+| **Resolution** | 384 × 384 center-cropped |
+| **Pixel Range** | 0–255 (unnormalized) |
+
+## Preprocessing Pipeline (V-JEPA Canonical)
+
+The preprocessing exactly follows Meta's V-JEPA training protocol:
+
+### 1. Frame Rate Downsampling
+- Source videos are downsampled to **4 FPS** using FFmpeg's  filter.
+- The  mode exactly matches Meta's  frame selection.
+
+### 2. Shorter-Side Scaling (Aspect-Ratio Preserving)
+- The shorter side of each frame is scaled to **438 pixels** while preserving the original aspect ratio.
+- This value is derived from the canonical V-JEPA proportion: .
+- For 224×224 crops, V-JEPA scales the shorter side to 256px. We scale proportionally for 384×384 crops.
+
+### 3. Center Cropping
+- A **384 × 384** center crop is taken from each scaled frame.
+- FFmpeg's  filter defaults to center cropping when no x/y offsets are specified.
+
+### 4. Tubelet Chunking
+- The full decoded frame sequence is divided into non-overlapping chunks of exactly **64 frames**.
+- Incomplete trailing chunks (fewer than 64 frames) are discarded.
+
+### FFmpeg Command Used
+
+
+## Directory Structure
+
+Files are partitioned across multiple root directories (, , , ...) to comply with HuggingFace's 10,000 files per directory limit.
+
+
+
+## Usage
+
+
+
+## Source Dataset
+
+- **Source**: [builddotai/Egocentric-10K](https://huggingface.co/datasets/builddotai/Egocentric-10K)
+- **Content**: Egocentric (first-person) video data from industrial/factory environments
+- **Factories**: factory_001 through factory_050
+
+## Hardware Used for Extraction
+
+- **GPU**: NVIDIA A40 (46 GB VRAM, 2 NVDEC chips)
+- **CPU**: 128 cores
+- **Decoding**: Hardware-accelerated NVDEC via FFmpeg with 10 concurrent decode threads