litert-community/MobileNet-v3-small

MobileNet V3 Small

MobileNet V3 model pre-trained on ImageNet-1k at resolution 224x224. Originally introduced by Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, Quoc V. Le, and Hartwig Adam in the paper, Searching for MobileNetV3.

Model description

The model was converted from a checkpoint from PyTorch Vision.

The original model has:
acc@1 (on ImageNet-1K): 67.668%
acc@5 (on ImageNet-1K): 87.402%
num_params: 2,542,856

The license information of the original model was missing.

Static INT8 status: work in progress

Static INT8 quantization is not available because naive post-training quantization causes severe classification accuracy degradation.

Naive post-training quantization of this MobileNetV3 Small checkpoint can lose important activation detail: channels with very different ranges share one INT8 activation scale, and the first depthwise convolution amplifies the resulting rounding error. Successful conversion or NPU compilation does not establish acceptable accuracy.

We are working on a validated solution. In the meantime, please use a larger model, such as MobileNetV3 Large, when you need static INT8 inference. No validated static INT8 MobileNetV3 Small checkpoint is currently available in this repository. The FP32 and weight-only INT8 variants listed below are available.

Available model files

File	Description
mobilenet_v3_small.tflite	Full precision LiteRT/TFLite model.
mobilenet_v3_small_weight_only_wi8_afp32.tflite	Weight-only INT8 model with FP32 activations.
mobilenet_v3_small_Google_Tensor_G5_apply_plugin.tflite	AOT-compiled artifact for the Google Tensor G5 target.

Quantized variant

mobilenet_v3_small_weight_only_wi8_afp32.tflite is a weight-only int8 quantization of the same weights (about 3.7x smaller than float32). Weight-only quantization is used instead of dynamic-range quantization because MobileNetV3's SE and hard-swish layers are sensitive to activation quantization; in a spot check against the float model the weight-only file keeps the top-1 predictions on real photos with a minimum logit correlation of 0.991.

Compatibility

File	CPU	GPU	NPU
mobilenet_v3_small.tflite	Supported	Supported	N/A

For GPU execution, explicitly select FP32 precision.

Intended uses & limitations

The model files were converted from pretrained weights from PyTorch Vision. The models may have their own licenses or terms and conditions derived from PyTorch Vision and the dataset used for training. It is your responsibility to determine whether you have permission to use the models for your use case.

How to Use

​​1. Install Dependencies Ensure your Python environment is set up with the required libraries. Run the following command in your terminal:

pip install numpy Pillow huggingface_hub ai-edge-litert

2. Prepare Your Image The script expects an image file to analyze. Make sure you have an image (e.g., cat.jpg or car.png) saved in the same working directory as your script.

3. Save the Script Create a new file named classify.py, paste the script below into it, and save the file:

#!/usr/bin/env python3
import argparse, json
import numpy as np
from PIL import Image
from huggingface_hub import hf_hub_download
from ai_edge_litert.compiled_model import CompiledModel

def preprocess(img: Image.Image) -> np.ndarray:
   img = img.convert("RGB")
   w, h = img.size
   s = 256
   if w < h:
       img = img.resize((s, int(h * s / w)), Image.BILINEAR)
   else:
       img = img.resize((int(w * s / h), s), Image.BILINEAR)
   left = int(round((img.size[0] - 224) / 2.0))
   top = int(round((img.size[1] - 224) / 2.0))
   img = img.crop((left, top, left + 224, top + 224))

   x = np.asarray(img, dtype=np.float32) / 255.0
   x = (x - np.array([0.485, 0.456, 0.406], dtype=np.float32)) / np.array(
       [0.229, 0.224, 0.225], dtype=np.float32
   )
   return np.ascontiguousarray(x.transpose(2, 0, 1)[None])

def main():
   ap = argparse.ArgumentParser()
   ap.add_argument("--image", required=True)
   args = ap.parse_args()

   model_path = hf_hub_download("litert-community/MobileNet-v3-small", "mobilenet_v3_small.tflite")
   labels_path = hf_hub_download(
       "huggingface/label-files", "imagenet-1k-id2label.json", repo_type="dataset"
   )
   with open(labels_path, "r", encoding="utf-8") as f:
       id2label = {int(k): v for k, v in json.load(f).items()}

   img = Image.open(args.image)
   x = preprocess(img)

   model = CompiledModel.from_file(model_path)
   inp = model.create_input_buffers(0)
   out = model.create_output_buffers(0)

   inp[0].write(x)
   model.run_by_index(0, inp, out)

   req = model.get_output_buffer_requirements(0, 0)
   y = out[0].read(req["buffer_size"] // np.dtype(np.float32).itemsize, np.float32)

   pred = int(np.argmax(y))
   label = id2label.get(pred, f"class_{pred}")

   print(f"Top-1 class index: {pred}")
   print(f"Top-1 label: {label}")
if __name__ == "__main__":
   main()

4. Execute the Python Script Run the below command:

python classify.py --image cat.jpg

BibTeX entry and citation info

@article{DBLP:journals/corr/abs-1905-02244,
  author       = {Andrew Howard and
                  Mark Sandler and
                  Grace Chu and
                  Liang{-}Chieh Chen and
                  Bo Chen and
                  Mingxing Tan and
                  Weijun Wang and
                  Yukun Zhu and
                  Ruoming Pang and
                  Vijay Vasudevan and
                  Quoc V. Le and
                  Hartwig Adam},
  title        = {Searching for MobileNetV3},
  journal      = {CoRR},
  volume       = {abs/1905.02244},
  year         = {2019},
  url          = {http://arxiv.org/abs/1905.02244},
  eprinttype    = {arXiv},
  eprint       = {1905.02244},
  timestamp    = {Thu, 27 May 2021 16:20:51 +0200},
  biburl       = {https://dblp.org/rec/journals/corr/abs-1905-02244.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}