--- pipeline_tag: image-classification license: cc-by-nc-4.0 base_model: timm/convnextv2_tiny.fcmae_ft_in1k library_name: kerasformers tags: - keras - kerasformers - image-classification - convnextv2 - backbone - arxiv:2301.00808 - pytorch - jax - tf --- ## ***See [our collection](https://huggingface.co/collections/kerasformers/convnext-v2-6a6bd292f10bf60554146f7b) for all versions of ConvNeXt-V2.*** # Run ConvNeXt-V2 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-ConvNeXt--V2-blue)](https://imvision12.github.io/KerasFormers/classification_backbones/) [![Collection](https://img.shields.io/badge/HF-ConvNeXt--V2%20collection-yellow)](https://huggingface.co/collections/kerasformers/convnext-v2-6a6bd292f10bf60554146f7b) # kerasformers/convnextv2_tiny_fcmae_ft_in1k Paper: [ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders (arXiv:2301.00808)](https://arxiv.org/abs/2301.00808) · [HF Papers](https://huggingface.co/papers/2301.00808) ConvNeXt V2 adds Global Response Normalization and FCMAE pretraining. Same classifier / backbone split as ConvNeXt. For more details on the model, please go to the upstream [model card](https://huggingface.co/timm/convnextv2_tiny.fcmae_ft_in1k). Pure-**Keras 3** conversion of [`timm/convnextv2_tiny.fcmae_ft_in1k`](https://huggingface.co/timm/convnextv2_tiny.fcmae_ft_in1k) for [kerasformers](https://github.com/IMvision12/KerasFormers). One implementation runs unmodified on **TensorFlow / Torch / JAX**. This is an **image-classification / backbone** checkpoint (`ConvNeXtV2ImageClassify` / `ConvNeXtV2Model`). ## ✨ Quick start ```python import os os.environ["KERAS_BACKEND"] = "torch" # or "jax" / "tensorflow" from PIL import Image import numpy as np from kerasformers.models.convnextv2 import ConvNeXtV2ImageClassify, ConvNeXtV2Model model = ConvNeXtV2ImageClassify.from_weights("kerasformers/convnextv2_tiny_fcmae_ft_in1k") backbone = ConvNeXtV2Model.from_weights( "kerasformers/convnextv2_tiny_fcmae_ft_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 ConvNeXt-V2 variant the same way with `from_weights("kerasformers/")`: | Variant | Hub | |---|---| | `convnextv2_atto_fcmae_ft_in1k` | [`kerasformers/convnextv2_atto_fcmae_ft_in1k`](https://huggingface.co/kerasformers/convnextv2_atto_fcmae_ft_in1k) | | `convnextv2_base_fcmae_ft_in1k` | [`kerasformers/convnextv2_base_fcmae_ft_in1k`](https://huggingface.co/kerasformers/convnextv2_base_fcmae_ft_in1k) | | `convnextv2_base_fcmae_ft_in22k_in1k` | [`kerasformers/convnextv2_base_fcmae_ft_in22k_in1k`](https://huggingface.co/kerasformers/convnextv2_base_fcmae_ft_in22k_in1k) | | `convnextv2_base_fcmae_ft_in22k_in1k_384` | [`kerasformers/convnextv2_base_fcmae_ft_in22k_in1k_384`](https://huggingface.co/kerasformers/convnextv2_base_fcmae_ft_in22k_in1k_384) | | `convnextv2_femto_fcmae_ft_in1k` | [`kerasformers/convnextv2_femto_fcmae_ft_in1k`](https://huggingface.co/kerasformers/convnextv2_femto_fcmae_ft_in1k) | | `convnextv2_huge_fcmae_ft_in1k` | [`kerasformers/convnextv2_huge_fcmae_ft_in1k`](https://huggingface.co/kerasformers/convnextv2_huge_fcmae_ft_in1k) | | `convnextv2_huge_fcmae_ft_in22k_in1k_384` | [`kerasformers/convnextv2_huge_fcmae_ft_in22k_in1k_384`](https://huggingface.co/kerasformers/convnextv2_huge_fcmae_ft_in22k_in1k_384) | | `convnextv2_huge_fcmae_ft_in22k_in1k_512` | [`kerasformers/convnextv2_huge_fcmae_ft_in22k_in1k_512`](https://huggingface.co/kerasformers/convnextv2_huge_fcmae_ft_in22k_in1k_512) | | `convnextv2_large_fcmae_ft_in1k` | [`kerasformers/convnextv2_large_fcmae_ft_in1k`](https://huggingface.co/kerasformers/convnextv2_large_fcmae_ft_in1k) | | `convnextv2_large_fcmae_ft_in22k_in1k` | [`kerasformers/convnextv2_large_fcmae_ft_in22k_in1k`](https://huggingface.co/kerasformers/convnextv2_large_fcmae_ft_in22k_in1k) | | `convnextv2_large_fcmae_ft_in22k_in1k_384` | [`kerasformers/convnextv2_large_fcmae_ft_in22k_in1k_384`](https://huggingface.co/kerasformers/convnextv2_large_fcmae_ft_in22k_in1k_384) | | `convnextv2_nano_fcmae_ft_in1k` | [`kerasformers/convnextv2_nano_fcmae_ft_in1k`](https://huggingface.co/kerasformers/convnextv2_nano_fcmae_ft_in1k) | | `convnextv2_nano_fcmae_ft_in22k_in1k` | [`kerasformers/convnextv2_nano_fcmae_ft_in22k_in1k`](https://huggingface.co/kerasformers/convnextv2_nano_fcmae_ft_in22k_in1k) | | `convnextv2_nano_fcmae_ft_in22k_in1k_384` | [`kerasformers/convnextv2_nano_fcmae_ft_in22k_in1k_384`](https://huggingface.co/kerasformers/convnextv2_nano_fcmae_ft_in22k_in1k_384) | | `convnextv2_pico_fcmae_ft_in1k` | [`kerasformers/convnextv2_pico_fcmae_ft_in1k`](https://huggingface.co/kerasformers/convnextv2_pico_fcmae_ft_in1k) | | `convnextv2_tiny_fcmae_ft_in1k` | [`kerasformers/convnextv2_tiny_fcmae_ft_in1k`](https://huggingface.co/kerasformers/convnextv2_tiny_fcmae_ft_in1k) | | `convnextv2_tiny_fcmae_ft_in22k_in1k` | [`kerasformers/convnextv2_tiny_fcmae_ft_in22k_in1k`](https://huggingface.co/kerasformers/convnextv2_tiny_fcmae_ft_in22k_in1k) | | `convnextv2_tiny_fcmae_ft_in22k_in1k_384` | [`kerasformers/convnextv2_tiny_fcmae_ft_in22k_in1k_384`](https://huggingface.co/kerasformers/convnextv2_tiny_fcmae_ft_in22k_in1k_384) | ## Tips - Set `KERAS_BACKEND` **before** importing Keras / kerasformers. - `ConvNeXtV2ImageClassify` returns class logits; `ConvNeXtV2Model` returns features (`as_backbone=True` for multi-scale stages). - See [docs](https://imvision12.github.io/KerasFormers/classification_backbones/) and [Loading Weights](https://imvision12.github.io/KerasFormers/loading_weights/). - Upstream / timm checkpoints: `ConvNeXtV2ImageClassify.from_weights("hf:timm/convnextv2_tiny.fcmae_ft_in1k")`. ## Special Thanks A huge thank you to the ConvNeXt-V2 authors and the timm / Hub communities for creating and releasing these models. License: see YAML `license` (usually matches the upstream checkpoint).