zeromodels/cait_m48_448_fb_dist_in1k

See our collection for all versions of CaiT.

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

kerasformers/cait_m48_448_fb_dist_in1k

Paper: Going deeper with Image Transformers (arXiv:2103.17239) · HF Papers

CaiT (Class-Attention in Image Transformers) deepens ViTs with class-attention layers. Distilled ImageNet variants included.

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

Pure-Keras 3 conversion of timm/cait_m48_448.fb_dist_in1k for kerasformers. One implementation runs unmodified on TensorFlow / Torch / JAX.

This is an image-classification / backbone checkpoint (CaiTImageClassify / CaiTModel).

✨ Quick start

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

from PIL import Image
import numpy as np
from kerasformers.models.cait import CaiTImageClassify, CaiTModel

model = CaiTImageClassify.from_weights("kerasformers/cait_m48_448_fb_dist_in1k")
backbone = CaiTModel.from_weights(
    "kerasformers/cait_m48_448_fb_dist_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 CaiT variant the same way with from_weights("kerasformers/<variant>"):

Variant	Hub
cait_m36_384_fb_dist_in1k	kerasformers/cait_m36_384_fb_dist_in1k
cait_m48_448_fb_dist_in1k	kerasformers/cait_m48_448_fb_dist_in1k
cait_s24_224_fb_dist_in1k	kerasformers/cait_s24_224_fb_dist_in1k
cait_s24_384_fb_dist_in1k	kerasformers/cait_s24_384_fb_dist_in1k
cait_s36_384_fb_dist_in1k	kerasformers/cait_s36_384_fb_dist_in1k
cait_xs24_384_fb_dist_in1k	kerasformers/cait_xs24_384_fb_dist_in1k
cait_xxs24_224_fb_dist_in1k	kerasformers/cait_xxs24_224_fb_dist_in1k
cait_xxs24_384_fb_dist_in1k	kerasformers/cait_xxs24_384_fb_dist_in1k
cait_xxs36_224_fb_dist_in1k	kerasformers/cait_xxs36_224_fb_dist_in1k
cait_xxs36_384_fb_dist_in1k	kerasformers/cait_xxs36_384_fb_dist_in1k

Tips

• Set KERAS_BACKEND before importing Keras / kerasformers.
• CaiTImageClassify returns class logits; CaiTModel returns features (as_backbone=True for multi-scale stages).
• See docs and Loading Weights.
• Upstream / timm checkpoints: CaiTImageClassify.from_weights("hf:timm/cait_m48_448.fb_dist_in1k").

Special Thanks

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

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