--- pipeline_tag: image-classification license: apache-2.0 base_model: timm/maxvit_tiny_tf_224.in1k library_name: zeromodels tags: - keras - zeromodels - image-classification - maxvit - backbone - arxiv:2204.01697 - pytorch - jax - tf --- ## ***See [our collection](https://huggingface.co/collections/zeromodels/maxvit-6a8eaee0e62bbd085eb81b76) for all versions of MaxViT.*** # Run MaxViT with Keras 3: JAX, PyTorch, or TensorFlow [![GitHub](https://img.shields.io/badge/GitHub-ZeroModels-black?logo=github)](https://github.com/IMvision12/ZeroModels) [![Docs](https://img.shields.io/badge/Docs-MaxViT-blue)](https://imvision12.github.io/ZeroModels/classification_backbones/) [![Collection](https://img.shields.io/badge/HF-MaxViT%20collection-yellow)](https://huggingface.co/collections/zeromodels/maxvit-6a8eaee0e62bbd085eb81b76) # zeromodels/maxvit_tiny_tf_224_in1k Paper: [MaxViT: Multi-Axis Vision Transformer (arXiv:2204.01697)](https://arxiv.org/abs/2204.01697) · [HF Papers](https://huggingface.co/papers/2204.01697) MaxViT combines blocked local and dilated global attention (multi-axis) in a hierarchical CNN/Transformer hybrid. For more details on the model, please go to the upstream [model card](https://huggingface.co/timm/maxvit_tiny_tf_224.in1k). Pure-**Keras 3** conversion of [`timm/maxvit_tiny_tf_224.in1k`](https://huggingface.co/timm/maxvit_tiny_tf_224.in1k) for [zeromodels](https://github.com/IMvision12/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**. This is an **image-classification / backbone** checkpoint (`MaxViTImageClassify` / `MaxViTModel`). ## ✨ Quick start ```python import os os.environ["KERAS_BACKEND"] = "torch" # or "jax" / "tensorflow" from PIL import Image import numpy as np from zeromodels.models.maxvit import MaxViTImageClassify, MaxViTModel model = MaxViTImageClassify.from_weights("zeromodels/maxvit_tiny_tf_224_in1k") backbone = MaxViTModel.from_weights( "zeromodels/maxvit_tiny_tf_224_in1k", as_backbone=True ) image = Image.open("your_image.jpg").convert("RGB") image = image.resize((224, 224)) x = np.asarray(image, dtype="float32")[None] # (1, H, W, 3) print(model(x).shape) # (1, num_classes) feats = backbone(x) print(len(feats), [tuple(f.shape) for f in feats]) ``` Load any MaxViT variant the same way with `from_weights("zeromodels/")`: | Variant | Hub | |---|---| | `maxvit_base_tf_224_in1k` | [`zeromodels/maxvit_base_tf_224_in1k`](https://huggingface.co/zeromodels/maxvit_base_tf_224_in1k) | | `maxvit_base_tf_224_in21k` | [`zeromodels/maxvit_base_tf_224_in21k`](https://huggingface.co/zeromodels/maxvit_base_tf_224_in21k) | | `maxvit_base_tf_384_in1k` | [`zeromodels/maxvit_base_tf_384_in1k`](https://huggingface.co/zeromodels/maxvit_base_tf_384_in1k) | | `maxvit_base_tf_384_in21k_ft_in1k` | [`zeromodels/maxvit_base_tf_384_in21k_ft_in1k`](https://huggingface.co/zeromodels/maxvit_base_tf_384_in21k_ft_in1k) | | `maxvit_base_tf_512_in1k` | [`zeromodels/maxvit_base_tf_512_in1k`](https://huggingface.co/zeromodels/maxvit_base_tf_512_in1k) | | `maxvit_base_tf_512_in21k_ft_in1k` | [`zeromodels/maxvit_base_tf_512_in21k_ft_in1k`](https://huggingface.co/zeromodels/maxvit_base_tf_512_in21k_ft_in1k) | | `maxvit_large_tf_224_in1k` | [`zeromodels/maxvit_large_tf_224_in1k`](https://huggingface.co/zeromodels/maxvit_large_tf_224_in1k) | | `maxvit_large_tf_224_in21k` | [`zeromodels/maxvit_large_tf_224_in21k`](https://huggingface.co/zeromodels/maxvit_large_tf_224_in21k) | | `maxvit_large_tf_384_in1k` | [`zeromodels/maxvit_large_tf_384_in1k`](https://huggingface.co/zeromodels/maxvit_large_tf_384_in1k) | | `maxvit_large_tf_384_in21k_ft_in1k` | [`zeromodels/maxvit_large_tf_384_in21k_ft_in1k`](https://huggingface.co/zeromodels/maxvit_large_tf_384_in21k_ft_in1k) | | `maxvit_large_tf_512_in1k` | [`zeromodels/maxvit_large_tf_512_in1k`](https://huggingface.co/zeromodels/maxvit_large_tf_512_in1k) | | `maxvit_large_tf_512_in21k_ft_in1k` | [`zeromodels/maxvit_large_tf_512_in21k_ft_in1k`](https://huggingface.co/zeromodels/maxvit_large_tf_512_in21k_ft_in1k) | | `maxvit_small_tf_224_in1k` | [`zeromodels/maxvit_small_tf_224_in1k`](https://huggingface.co/zeromodels/maxvit_small_tf_224_in1k) | | `maxvit_small_tf_384_in1k` | [`zeromodels/maxvit_small_tf_384_in1k`](https://huggingface.co/zeromodels/maxvit_small_tf_384_in1k) | | `maxvit_small_tf_512_in1k` | [`zeromodels/maxvit_small_tf_512_in1k`](https://huggingface.co/zeromodels/maxvit_small_tf_512_in1k) | | `maxvit_tiny_tf_224_in1k` | [`zeromodels/maxvit_tiny_tf_224_in1k`](https://huggingface.co/zeromodels/maxvit_tiny_tf_224_in1k) | | `maxvit_tiny_tf_384_in1k` | [`zeromodels/maxvit_tiny_tf_384_in1k`](https://huggingface.co/zeromodels/maxvit_tiny_tf_384_in1k) | | `maxvit_tiny_tf_512_in1k` | [`zeromodels/maxvit_tiny_tf_512_in1k`](https://huggingface.co/zeromodels/maxvit_tiny_tf_512_in1k) | | `maxvit_xlarge_tf_224_in21k` | [`zeromodels/maxvit_xlarge_tf_224_in21k`](https://huggingface.co/zeromodels/maxvit_xlarge_tf_224_in21k) | | `maxvit_xlarge_tf_384_in21k_ft_in1k` | [`zeromodels/maxvit_xlarge_tf_384_in21k_ft_in1k`](https://huggingface.co/zeromodels/maxvit_xlarge_tf_384_in21k_ft_in1k) | | `maxvit_xlarge_tf_512_in21k_ft_in1k` | [`zeromodels/maxvit_xlarge_tf_512_in21k_ft_in1k`](https://huggingface.co/zeromodels/maxvit_xlarge_tf_512_in21k_ft_in1k) | ## Tips - Set `KERAS_BACKEND` **before** importing Keras / zeromodels. - `MaxViTImageClassify` returns class logits; `MaxViTModel` returns features (`as_backbone=True` for multi-scale stages). - See [docs](https://imvision12.github.io/ZeroModels/classification_backbones/) and [Loading Weights](https://imvision12.github.io/ZeroModels/loading_weights/). - Upstream / timm checkpoints: `MaxViTImageClassify.from_weights("hf:timm/maxvit_tiny_tf_224.in1k")`. ## Special Thanks A huge thank you to the MaxViT authors and the timm / Hub communities for creating and releasing these models. License: see YAML `license` (usually matches the upstream checkpoint).