zeromodels/flexivit_base_1200ep_in1k

See our collection for all versions of FlexiViT.

Run FlexiViT with Keras 3: JAX, PyTorch, or TensorFlow

zeromodels/flexivit_base_1200ep_in1k

Paper: FlexiViT: One Model for All Patch Sizes (arXiv:2212.08013) · HF Papers

FlexiViT trains one ViT that transfers across patch sizes at inference. Classifier or token backbone.

For more details on the model, please go to the upstream model card.

Pure-Keras 3 conversion of timm/flexivit_base.1200ep_in1k for zeromodels. One implementation runs unmodified on TensorFlow / Torch / JAX.

This is an image-classification / backbone checkpoint (FlexiViTImageClassify / FlexiViTModel).

✨ Quick start

import os

os.environ["KERAS_BACKEND"] = "torch"  # or "jax" / "tensorflow"

from PIL import Image
from zeromodels.models.flexivit import FlexiViTImageClassify, FlexiViTModel, FlexiViTImageProcessor

model = FlexiViTImageClassify.from_weights("zeromodels/flexivit_base_1200ep_in1k")
processor = FlexiViTImageProcessor.from_weights("zeromodels/flexivit_base_1200ep_in1k")

image = Image.open("your_image.jpg").convert("RGB")
pixels = processor(image)  # resize + normalize (normalization lives in the processor)
logits = model(pixels, training=False)
print(logits.shape)  # (1, num_classes)

# Feature extraction: the backbone without the classifier head
backbone = FlexiViTModel.from_weights("zeromodels/flexivit_base_1200ep_in1k", as_backbone=True)
features = backbone(pixels, training=False)

Load any FlexiViT variant the same way with from_weights("zeromodels/<variant>"):

Variant	Hub
flexivit_base_1000ep_in21k	zeromodels/flexivit_base_1000ep_in21k
flexivit_base_1200ep_in1k	zeromodels/flexivit_base_1200ep_in1k
flexivit_base_300ep_in1k	zeromodels/flexivit_base_300ep_in1k
flexivit_base_300ep_in21k	zeromodels/flexivit_base_300ep_in21k
flexivit_large_1200ep_in1k	zeromodels/flexivit_large_1200ep_in1k
flexivit_large_300ep_in1k	zeromodels/flexivit_large_300ep_in1k
flexivit_large_600ep_in1k	zeromodels/flexivit_large_600ep_in1k
flexivit_small_1200ep_in1k	zeromodels/flexivit_small_1200ep_in1k
flexivit_small_300ep_in1k	zeromodels/flexivit_small_300ep_in1k
flexivit_small_600ep_in1k	zeromodels/flexivit_small_600ep_in1k

Tips

• Set KERAS_BACKEND before importing Keras / zeromodels.
• FlexiViTImageClassify returns class logits; FlexiViTModel returns features (as_backbone=True for multi-scale stages).
• See docs and Loading Weights.
• Upstream / timm checkpoints: FlexiViTImageClassify.from_weights("hf:timm/flexivit_base.1200ep_in1k").

Special Thanks

A huge thank you to the FlexiViT authors and the timm / Hub communities for creating and releasing these models.

License: see YAML license (usually matches the upstream checkpoint).