Update README.md
Browse files
README.md
CHANGED
|
@@ -3,4 +3,63 @@ license: other
|
|
| 3 |
license_name: apple-ascl
|
| 4 |
license_link: LICENSE
|
| 5 |
library_name: mobileclip
|
| 6 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
license_name: apple-ascl
|
| 4 |
license_link: LICENSE
|
| 5 |
library_name: mobileclip
|
| 6 |
+
---
|
| 7 |
+
|
| 8 |
+
# MobileCLIP: Fast Image-Text Models through Multi-Modal Reinforced Training
|
| 9 |
+
|
| 10 |
+
MobileCLIP was introduced in [MobileCLIP: Fast Image-Text Models through Multi-Modal Reinforced Training
|
| 11 |
+
](https://arxiv.org/pdf/2311.17049.pdf) (CVPR 2024), by Pavan Kumar Anasosalu Vasu, Hadi Pouransari, Fartash Faghri, Raviteja Vemulapalli, Oncel Tuzel.
|
| 12 |
+
|
| 13 |
+
This repository contains the **MobileCLIP-B** checkpoint.
|
| 14 |
+
|
| 15 |
+

|
| 16 |
+
|
| 17 |
+
### Highlights
|
| 18 |
+
|
| 19 |
+
* Our smallest variant `MobileCLIP-S0` obtains similar zero-shot performance as [OpenAI](https://arxiv.org/abs/2103.00020)'s ViT-B/16 model while being 4.8x faster and 2.8x smaller.
|
| 20 |
+
* `MobileCLIP-S2` obtains better avg zero-shot performance than [SigLIP](https://arxiv.org/abs/2303.15343)'s ViT-B/16 model while being 2.3x faster and 2.1x smaller, and trained with 3x less seen samples.
|
| 21 |
+
* `MobileCLIP-B`(LT) attains zero-shot ImageNet performance of **77.2%** which is significantly better than recent works like [DFN](https://arxiv.org/abs/2309.17425) and [SigLIP](https://arxiv.org/abs/2303.15343) with similar architectures or even [OpenAI's ViT-L/14@336](https://arxiv.org/abs/2103.00020).
|
| 22 |
+
|
| 23 |
+
## Checkpoints
|
| 24 |
+
|
| 25 |
+
| Model | # Seen <BR>Samples (B) | # Params (M) <BR> (img + txt) | Latency (ms) <BR> (img + txt) | IN-1k Zero-Shot <BR> Top-1 Acc. (%) | Avg. Perf. (%) <BR> on 38 datasets |
|
| 26 |
+
|:----------------------------------------------------------|:----------------------:|:-----------------------------:|:-----------------------------:|:-----------------------------------:|:----------------------------------:|
|
| 27 |
+
| [MobileCLIP-S0](https://hf.co/pcuenq/MobileCLIP-S0) | 13 | 11.4 + 42.4 | 1.5 + 1.6 | 67.8 | 58.1 |
|
| 28 |
+
| [MobileCLIP-S1](https://hf.co/pcuenq/MobileCLIP-S1) | 13 | 21.5 + 63.4 | 2.5 + 3.3 | 72.6 | 61.3 |
|
| 29 |
+
| [MobileCLIP-S2](https://hf.co/pcuenq/MobileCLIP-S2) | 13 | 35.7 + 63.4 | 3.6 + 3.3 | 74.4 | 63.7 |
|
| 30 |
+
| [MobileCLIP-B](https://hf.co/pcuenq/MobileCLIP-B) | 13 | 86.3 + 63.4 | 10.4 + 3.3 | 76.8 | 65.2 |
|
| 31 |
+
| [MobileCLIP-B (LT)](https://hf.co/pcuenq/MobileCLIP-B-LT) | 36 | 86.3 + 63.4 | 10.4 + 3.3 | 77.2 | 65.8 |
|
| 32 |
+
|
| 33 |
+
## How to Use
|
| 34 |
+
|
| 35 |
+
First, download the desired checkpoint visiting one of the links in the table above, then click the `Files and versions` tab, and download the PyTorch checkpoint.
|
| 36 |
+
For programmatic downloading, if you have `huggingface_hub` installed, you can also run:
|
| 37 |
+
|
| 38 |
+
```
|
| 39 |
+
huggingface-cli download pcuenq/MobileCLIP-B
|
| 40 |
+
```
|
| 41 |
+
|
| 42 |
+
Then, install [`ml-mobileclip`](https://github.com/apple/ml-mobileclip) by following the instructions in the repo. It uses an API similar to [`open_clip`'s](https://github.com/mlfoundations/open_clip).
|
| 43 |
+
You can run inference with a code snippet like the following:
|
| 44 |
+
|
| 45 |
+
```py
|
| 46 |
+
import torch
|
| 47 |
+
from PIL import Image
|
| 48 |
+
import mobileclip
|
| 49 |
+
|
| 50 |
+
model, _, preprocess = mobileclip.create_model_and_transforms('mobileclip_s0', pretrained='/path/to/mobileclip_s0.pt')
|
| 51 |
+
tokenizer = mobileclip.get_tokenizer('mobileclip_s0')
|
| 52 |
+
|
| 53 |
+
image = preprocess(Image.open("docs/fig_accuracy_latency.png").convert('RGB')).unsqueeze(0)
|
| 54 |
+
text = tokenizer(["a diagram", "a dog", "a cat"])
|
| 55 |
+
|
| 56 |
+
with torch.no_grad(), torch.cuda.amp.autocast():
|
| 57 |
+
image_features = model.encode_image(image)
|
| 58 |
+
text_features = model.encode_text(text)
|
| 59 |
+
image_features /= image_features.norm(dim=-1, keepdim=True)
|
| 60 |
+
text_features /= text_features.norm(dim=-1, keepdim=True)
|
| 61 |
+
|
| 62 |
+
text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1)
|
| 63 |
+
|
| 64 |
+
print("Label probs:", text_probs)
|
| 65 |
+
```
|