metadata
tags:
- image-classification
- timm
- MobileNetV4
license: apache-2.0
datasets:
- imagenet-1k
pipeline_tag: image-classification
Model card for MobileNetV4_Conv_Medium_TFLite_256
A MobileNet-V4 image classification model. Trained on ImageNet-1k by Ross Wightman.
Converted to TFLite Float32 & Float16 formats by Youssef Boulaouane.
Note from original pytorch model: This is a customization with blur pooling for downsample convs after the stem.
Model Details
- Pytorch Weights: https://huggingface.co/timm/mobilenetv4_conv_medium.e500_r256_in1k
- Model Type: Image classification
- Model Stats:
- Params (M): 9.7
- GMACs: 1.1
- Activations (M): 7.6
- Input Shape (1, 256, 256, 3)
- Dataset: ImageNet-1k
- Papers:
- MobileNetV4 -- Universal Models for the Mobile Ecosystem: https://arxiv.org/abs/2404.10518
- PyTorch Image Models: https://github.com/huggingface/pytorch-image-models
- Original: https://github.com/tensorflow/models/tree/master/official/vision
Model Usage
Image Classification in Python
import numpy as np
import tensorflow as tf
from PIL import Image
# Load label file
with open('imagenet_classes.txt', 'r') as file:
lines = file.readlines()
index_to_label = {index: line.strip() for index, line in enumerate(lines)}
# Initialize interpreter and IO details
tfl_model = tf.lite.Interpreter(model_path=tf_model_path)
tfl_model.allocate_tensors()
input_details = tfl_model.get_input_details()
output_details = tfl_model.get_output_details()
# Load and preprocess the image
image = Image.open(image_path).resize((256, 256), Image.BICUBIC)
image = np.array(image, dtype=np.float32)
mean = np.array([0.485, 0.456, 0.406], dtype=np.float32)
std = np.array([0.229, 0.224, 0.225], dtype=np.float32)
image = (image / 255.0 - mean) / std
image = np.expand_dims(image, axis=-1)
image = np.rollaxis(image, 3)
# Inference and postprocessing
input = input_details[0]
tfl_model.set_tensor(input["index"], image)
tfl_model.invoke()
tfl_output = tfl_model.get_tensor(output_details[0]["index"])
tfl_output_tensor = tf.convert_to_tensor(tfl_output)
tfl_softmax_output = tf.nn.softmax(tfl_output_tensor, axis=1)
tfl_top5_probs, tfl_top5_indices = tf.math.top_k(tfl_softmax_output, k=5)
# Get the top5 class labels and probabilities
tfl_probs_list = tfl_top5_probs[0].numpy().tolist()
tfl_index_list = tfl_top5_indices[0].numpy().tolist()
for index, prob in zip(tfl_index_list, tfl_probs_list):
print(f"{index_to_label[index]}: {round(prob*100, 2)}%")
Deployment on Mobile
Refer to guides available here: https://ai.google.dev/edge/lite/inference
Citation
@article{qin2024mobilenetv4,
title={MobileNetV4-Universal Models for the Mobile Ecosystem},
author={Qin, Danfeng and Leichner, Chas and Delakis, Manolis and Fornoni, Marco and Luo, Shixin and Yang, Fan and Wang, Weijun and Banbury, Colby and Ye, Chengxi and Akin, Berkin and others},
journal={arXiv preprint arXiv:2404.10518},
year={2024}
}
@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}}
}