fix readme.md

README.md

@@ -1,22 +1,97 @@
 ---
 pipeline_tag: zero-shot-image-classification
 license: apache-2.0
+base_model: google/siglip2-large-patch16-384
 library_name: kerasformers
 tags:
 - keras
 - kerasformers
 - siglip2
-- tf
-- jax
+- zero-shot-image-classification
+- vision
+- arxiv:2502.14786
 - pytorch
+- jax
+- tf
 ---
 
-# siglip2_large_p16_384 (Keras 3)
+## ***See [our collection](https://huggingface.co/collections/kerasformers/siglip2-6a6ab35c79c0333394386a90) for all versions of SigLIP 2.***
+
+# Run SigLIP 2 with Keras 3: JAX, PyTorch, or TensorFlow
+
+[![GitHub](https://img.shields.io/badge/GitHub-KerasFormers-black?logo=github)](https://github.com/IMvision12/KerasFormers) [![Docs](https://img.shields.io/badge/Docs-SigLIP--2-blue)](https://imvision12.github.io/KerasFormers/siglip2/) [![Collection](https://img.shields.io/badge/HF-SigLIP--2%20collection-yellow)](https://huggingface.co/collections/kerasformers/siglip2-6a6ab35c79c0333394386a90)
+
+# kerasformers/siglip2_large_p16_384
+
+Paper: [SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features (arXiv:2502.14786)](https://arxiv.org/abs/2502.14786) · [HF Papers](https://huggingface.co/papers/2502.14786)
+
+SigLIP 2 keeps SigLIP's sigmoid loss and adds captioning-based pretraining, self-distillation, and masked prediction for stronger dense features. It uses a 256k multilingual Gemma vocabulary, so many languages work without a separate multilingual checkpoint.
+
+For more details on the model, please go to the upstream [model card](https://huggingface.co/google/siglip2-large-patch16-384).
+
+Pure-**Keras 3** conversion of [`google/siglip2-large-patch16-384`](https://huggingface.co/google/siglip2-large-patch16-384) for [kerasformers](https://github.com/IMvision12/KerasFormers). One implementation runs unmodified on **TensorFlow / Torch / JAX**.
 
-Pure-Keras 3 weights for [kerasformers](https://github.com/IMvision12/KerasFormers), mirrored from the GitHub release. License: `apache-2.0`.
+This is a **zero-shot image-text** checkpoint (`SigLIP2ZeroShotClassify`): pass image(s) and text prompts at inference time.
+
+## ✨ Quick start
 
 ```python
-from kerasformers.models.siglip2 import SigLIP2Model, SigLIP2Tokenizer
-model = SigLIP2Model.from_weights("siglip2_large_p16_384")
-tokenizer = SigLIP2Tokenizer.from_weights("siglip2_large_p16_384")
+import os
+os.environ["KERAS_BACKEND"] = "torch"  # or "jax" / "tensorflow"
+
+from kerasformers.models.siglip2 import (
+    SigLIP2Processor,
+    SigLIP2ZeroShotClassify,
+)
+
+processor = SigLIP2Processor.from_weights("kerasformers/siglip2_large_p16_384")
+model = SigLIP2ZeroShotClassify.from_weights("kerasformers/siglip2_large_p16_384")
+
+labels = [
+    "a photo of a cat",
+    "a photo of a dog",
+    "a photo of a car",
+    "a photo of a living room",
+]
+inputs = processor(text=labels, image_paths="your_image.jpg")
+output = model(
+    {
+        "images": inputs["images"],
+        "token_ids": inputs["input_ids"],
+    }
+)
+print(output["image_logits"].shape)
 ```
+
+Load any SigLIP 2 variant the same way with `from_weights("kerasformers/<variant>")`:
+
+| Variant | Hub |
+|---|---|
+| `siglip2_base_p16_224` | [`kerasformers/siglip2_base_p16_224`](https://huggingface.co/kerasformers/siglip2_base_p16_224) |
+| `siglip2_base_p16_256` | [`kerasformers/siglip2_base_p16_256`](https://huggingface.co/kerasformers/siglip2_base_p16_256) |
+| `siglip2_base_p16_384` | [`kerasformers/siglip2_base_p16_384`](https://huggingface.co/kerasformers/siglip2_base_p16_384) |
+| `siglip2_base_p16_512` | [`kerasformers/siglip2_base_p16_512`](https://huggingface.co/kerasformers/siglip2_base_p16_512) |
+| `siglip2_base_p32_256` | [`kerasformers/siglip2_base_p32_256`](https://huggingface.co/kerasformers/siglip2_base_p32_256) |
+| `siglip2_large_p16_256` | [`kerasformers/siglip2_large_p16_256`](https://huggingface.co/kerasformers/siglip2_large_p16_256) |
+| `siglip2_large_p16_384` | [`kerasformers/siglip2_large_p16_384`](https://huggingface.co/kerasformers/siglip2_large_p16_384) |
+| `siglip2_large_p16_512` | [`kerasformers/siglip2_large_p16_512`](https://huggingface.co/kerasformers/siglip2_large_p16_512) |
+| `siglip2_so400m_p14_224` | [`kerasformers/siglip2_so400m_p14_224`](https://huggingface.co/kerasformers/siglip2_so400m_p14_224) |
+| `siglip2_so400m_p14_384` | [`kerasformers/siglip2_so400m_p14_384`](https://huggingface.co/kerasformers/siglip2_so400m_p14_384) |
+| `siglip2_so400m_p16_256` | [`kerasformers/siglip2_so400m_p16_256`](https://huggingface.co/kerasformers/siglip2_so400m_p16_256) |
+| `siglip2_so400m_p16_384` | [`kerasformers/siglip2_so400m_p16_384`](https://huggingface.co/kerasformers/siglip2_so400m_p16_384) |
+| `siglip2_so400m_p16_512` | [`kerasformers/siglip2_so400m_p16_512`](https://huggingface.co/kerasformers/siglip2_so400m_p16_512) |
+
+## Tips
+
+- Set `KERAS_BACKEND` **before** importing Keras / kerasformers.
+- Prefer `Processor.from_weights(...)` so image size and tokenizer match the variant.
+- Map processor `input_ids` to model `token_ids`. No padding mask is required.
+- Tokenizer is Gemma-based (multilingual); prefer `Processor.from_weights`.
+- See [SigLIP 2 docs](https://imvision12.github.io/KerasFormers/siglip2/) and [Loading Weights](https://imvision12.github.io/KerasFormers/loading_weights/).
+- Community / upstream safetensors still work via the `hf:` prefix, e.g. `SigLIP2ZeroShotClassify.from_weights("hf:google/siglip2-large-patch16-384")`.
+
+## Special Thanks
+
+A huge thank you to the Google SigLIP 2 authors for creating and releasing these models.
+
+License: Apache 2.0.