siriusdemon / pytorch-dex Goto Github PK

Pytorch implementation of DEX: Deep EXpectation of apparent age from a single image

Python 100.00%

gender-prediction age-prediction gender age

pytorch-dex's Introduction

DEX: Deep EXpectation of apparent age from a single image

This is a pytorch version of DEX. Refer to its Home Page for more details

You can refer to insight if you want a much smaller model but it uses mxnet instead of pytorch. I haven't convert it to pytorch yet.

Getting Started

A separate Python environment is recommended.

Python3.5+ (Python3.5, Python3.6 are tested)
Pytorch == 1.0
opencv4 (opencv3.4.5 is tested also)
numpy

install dependences using pip

pip3 install numpy opencv-python
pip3 install https://download.pytorch.org/whl/cpu/torch-1.0.1.post2-cp36-cp36m-linux_x86_64.whl
pip3 install torchvision (optional)

or install using conda

conda install opencv numpy
conda install pytorch-cpu torchvision-cpu -c pytorch

Usage

git clone https://github.com/siriusdemon/pytorch-DEX.git
cd pytorch-DEX
python demo.py path/to/image

Results

predict image: imgs/2.png
woman: 0.994, man: 0.006
age: 21.433

predict image: imgs/5.png
woman: 0.010, man: 0.990
age: 42.896

Installation

You can use dex as a separate Python package right now!

cd pytorch-DEX
pip install .

See demo.py for example.

Citation

@InProceedings{Rothe-ICCVW-2015,
  author = {Rasmus Rothe and Radu Timofte and Luc Van Gool},
  title = {DEX: Deep EXpectation of apparent age from a single image},
  booktitle = {IEEE International Conference on Computer Vision Workshops (ICCVW)},
  year = {2015},
  month = {December},
}

pytorch-dex's People

Contributors

Stargazers

Watchers

pytorch-dex's Issues

Pytorch Models

Hello!
I have a question, have you trained the model by yourself? or you somehow converted the original caffe model to Pytorch?

您好，我想在自己的数据上训练模型，怎么做呢？

您好，我想在自己的数据上训练模型。并且在网站https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/中没有找到相关的代码，请问您是在哪里看到的呢，能不能提供一个具体的地址。万分感谢！

some minor mistakes of the code.

Image Normalization according to imagenet_mean.binaryproto.
Age Estimation computing mistake.
Your code:

def expected_age(vector):
    res = [(i+1)*v for i, v in enumerate(vector)]
    return sum(res)

but in the paper:
$E(O)=\sum_{i=0}^{100} y_{i} o_{i}$

I have encountered a problem：
Traceback (most recent call last):
File "demo.py", line 5, in
dex.eval()
File "/home/zyd/Reed/DL/age-estimation/code/pytorch-DEX/dex/api.py", line 23, in _eval
gender_model.load_state_dict(torch.load(gender_model_path))
File "/home/zyd/anaconda3/envs/pytorch/lib/python3.5/site-packages/torch/serialization.py", line 367, in load
return _load(f, map_location, pickle_module)
File "/home/zyd/anaconda3/envs/pytorch/lib/python3.5/site-packages/torch/serialization.py", line 528, in _load
magic_number = pickle_module.load(f)
_pickle.UnpicklingError: invalid load key, 'v'.

How can I solve it?

Pre-trained on which dataset?

May I ask that this models are pre-trained on which dataset? IMDB-WIKI? or LAP?

How can I train the network using my own datasets?

I used your model file, but got an error of about 8-10 years old. I want to re-use the data set for training. What should I do?

Accuracy IMDB-dataset

I am a student who is trying to train the network. How could I accurately select the images in the dataset? I noticed that many photos do not correspond to the real case.

git-lfs data quota over

Hi,
when pulling the repo the following error shows:
Error downloading object: dex/pth/age_sd.pth (d754f06): Smudge error: Error downloading dex/pth/age_sd.pth (d754f06341792aecf1c5893269ae849892072532f161b63cd595d713897f5302): batch response: This repository is over its data quota. Purchase more data packs to restore access.
Can you maybe remove the lfs option and provide download links to the pretrained models?

about the loss function in training

hi, sorry to bother. I wanna use your written model to train my own datasets. noticed that the expected_age is calculated directly by the output of model. when doing training, is the loss function cross-entropy between the output of model and the sequence where 1 at the age and 0 otherwise?