maxvit_tiny_rw_224

maxvit_tiny_rw_224

Converted TIMM image classification model for LiteRT.

• Source architecture: maxvit_tiny_rw_224
• Source checkpoint: timm/maxvit_tiny_rw_224.sw_in1k
• FP32 file: model.tflite
• Static INT8 file: model_static_int8.tflite
• Input: NCHW tensor, shape [1, 3, 224, 224] (FP32 or INT8 according to the file)
• Output: ImageNet-1K logits, shape [1, 1000]

Compatibility

The NPU entry denotes compilation support; runtime accuracy is unverified.

model_static_int8.tflite uses INT8 inputs/outputs and convolution/FC quantization; attention, normalization and other operations remain FP32. Apply the source checkpoint’s preprocessing, then quantize the input and dequantize the output using the file’s tensor scales and zero points.

For GPU execution of model.tflite, select FP32 GPU precision.

Model Details

• Model Type: Image classification / feature backbone
• Model Stats:
  • Params (M): 29.1
  • GMACs: 5.1
  • Activations (M): 33.1
  • Image size: 224 x 224
• Papers:
  • MaxViT: Multi-Axis Vision Transformer: https://arxiv.org/abs/2204.01697
• Dataset: ImageNet-1k

Citation

@misc{rw2019timm,
  author = {Ross Wightman},
  title = {PyTorch Image Models},
  year = {2019},
  publisher = {GitHub},
  journal = {GitHub repository},
  doi = {10.5281/zenodo.4414861},
  howpublished = {\url{https://github.com/huggingface/pytorch-image-models}}
}

@article{tu2022maxvit,
  title={MaxViT: Multi-Axis Vision Transformer},
  author={Tu, Zhengzhong and Talebi, Hossein and Zhang, Han and Yang, Feng and Milanfar, Peyman and Bovik, Alan and Li, Yinxiao},
  journal={ECCV},
  year={2022},
}        

@article{dai2021coatnet,
  title={CoAtNet: Marrying Convolution and Attention for All Data Sizes},
  author={Dai, Zihang and Liu, Hanxiao and Le, Quoc V and Tan, Mingxing},
  journal={arXiv preprint arXiv:2106.04803},
  year={2021}
}