Update README.md
Browse files
README.md
CHANGED
|
@@ -2,8 +2,49 @@
|
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
|
|
|
|
| 5 |
|
| 6 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
|
| 8 |
> **Note:** This model supports native resolution input. For optimal performance:
|
| 9 |
> - **Image**: 448×448 resolution (pre-trained)
|
|
|
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
|
| 5 |
+
### Model Details
|
| 6 |
|
| 7 |
+
| Property | Value |
|
| 8 |
+
|----------|-------|
|
| 9 |
+
| **Model Type** | Vision Transformer (ViT) |
|
| 10 |
+
| **Architecture** | HEVC-Style Vision Transformer |
|
| 11 |
+
| **Hidden Size** | 1024 |
|
| 12 |
+
| **Intermediate Size** | 4096 |
|
| 13 |
+
| **Number of Layers** | 24 |
|
| 14 |
+
| **Number of Attention Heads** | 16 |
|
| 15 |
+
| **Patch Size** | 16 |
|
| 16 |
+
| **Image Resolution** | 448×448 (pre-trained) |
|
| 17 |
+
| **Video Resolution** | 224×224 with 256 tokens per frame |
|
| 18 |
+
| **Positional Encoding** | 3D RoPE (4:6:6 split for T:H:W) |
|
| 19 |
+
| **Normalization** | Layer Normalization |
|
| 20 |
+
| **Activation Function** | GELU |
|
| 21 |
+
| **Attention Implementation** | Flash Attention 2 |
|
| 22 |
+
| **License** | Apache 2.0 |
|
| 23 |
+
|
| 24 |
+
### Key Features
|
| 25 |
+
|
| 26 |
+
- **Codec-Style Patch Selection**: Instead of sampling sparse frames densely (all patches from few frames), OneVision Encoder samples dense frames sparsely (important patches from many frames).
|
| 27 |
+
- **3D Rotary Position Embedding**: Uses a 4:6:6 split for temporal, height, and width dimensions to capture spatiotemporal relationships.
|
| 28 |
+
- **Global Contrastive Learning**: Trained with a 2M concept bank for better-separated semantic clusters.
|
| 29 |
+
- **Native Resolution Support**: Supports native resolution input without tiling or cropping.
|
| 30 |
+
- **Flash Attention 2**: Efficient attention implementation for improved performance and memory efficiency.
|
| 31 |
+
|
| 32 |
+
### Intended Use
|
| 33 |
+
|
| 34 |
+
#### Primary Use Cases
|
| 35 |
+
|
| 36 |
+
- **Video Understanding**: Action recognition, video captioning, video question answering
|
| 37 |
+
- **Image Understanding**: Document understanding (DocVQA), chart understanding (ChartQA), OCR tasks
|
| 38 |
+
- **Vision-Language Models**: As the vision encoder backbone for multimodal large language models
|
| 39 |
+
|
| 40 |
+
#### Downstream Tasks
|
| 41 |
+
|
| 42 |
+
- Video benchmarks: MVBench, VideoMME, Perception Test
|
| 43 |
+
- Image understanding: DocVQA, ChartQA, OCRBench
|
| 44 |
+
- Action recognition: SSv2, UCF101, Kinetics
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
### Quick Start
|
| 48 |
|
| 49 |
> **Note:** This model supports native resolution input. For optimal performance:
|
| 50 |
> - **Image**: 448×448 resolution (pre-trained)
|