Clarify DINOv2-small is a frozen encoder, not a fine-tuned base model
Browse files
README.md
CHANGED
|
@@ -4,7 +4,6 @@ library_name: pytorch
|
|
| 4 |
pipeline_tag: image-to-text
|
| 5 |
datasets:
|
| 6 |
- jxie/coco_captions
|
| 7 |
-
base_model: facebook/dinov2-small
|
| 8 |
tags:
|
| 9 |
- image-captioning
|
| 10 |
- video
|
|
@@ -22,7 +21,10 @@ description the assistant can read, through MCP (Claude Desktop, Claude Code, Cu
|
|
| 22 |
VS Code, LM Studio and other MCP apps) or as plain text for any LLM. It runs locally, so
|
| 23 |
nothing is uploaded.
|
| 24 |
|
| 25 |
-
It's a **36M-parameter** image and video captioner, fast on a laptop CPU.
|
|
|
|
|
|
|
|
|
|
| 26 |
|
| 27 |
- **89% of Florence-2-base's COCO caption score at 1/6 the size**
|
| 28 |
- **8× faster on GPU, 75× faster on CPU** (54 ms per image on an Apple M5 CPU)
|
|
|
|
| 4 |
pipeline_tag: image-to-text
|
| 5 |
datasets:
|
| 6 |
- jxie/coco_captions
|
|
|
|
| 7 |
tags:
|
| 8 |
- image-captioning
|
| 9 |
- video
|
|
|
|
| 21 |
VS Code, LM Studio and other MCP apps) or as plain text for any LLM. It runs locally, so
|
| 22 |
nothing is uploaded.
|
| 23 |
|
| 24 |
+
It's a **36M-parameter** image and video captioner, fast on a laptop CPU. It uses Meta's
|
| 25 |
+
DINOv2-small as a frozen image encoder (its weights are unchanged, so this is not a
|
| 26 |
+
fine-tune of DINOv2); the adapter and caption decoder on top (14M parameters) were
|
| 27 |
+
trained from scratch on COCO.
|
| 28 |
|
| 29 |
- **89% of Florence-2-base's COCO caption score at 1/6 the size**
|
| 30 |
- **8× faster on GPU, 75× faster on CPU** (54 ms per image on an Apple M5 CPU)
|