onnxmodelzoo/emotion-ferplus-7

FER+ Emotion Recognition

Description

This model is a deep convolutional neural network for emotion recognition in faces.

Model

Model	Download	Download (with sample test data)	ONNX version	Opset version
Emotion FERPlus	34 MB	31 MB	1.0	2
Emotion FERPlus	34 MB	31 MB	1.2	7
Emotion FERPlus	34 MB	31 MB	1.3	8
Emotion FERPlus int8	19 MB	18 MB	1.14	12

Paper

"Training Deep Networks for Facial Expression Recognition with Crowd-Sourced Label Distribution" arXiv:1608.01041

Dataset

The model is trained on the FER+ annotations for the standard Emotion FER dataset, as described in the above paper.

Source

The model is trained in CNTK, using the cross entropy training mode. You can find the source code here.

Demo

Run Emotion_FERPlus in browser - implemented by ONNX.js with Emotion_FERPlus version 1.2

Inference

Input

The model expects input of the shape (Nx1x64x64), where N is the batch size.

Preprocessing

Given a path image_path to the image you would like to score:

import numpy as np
from PIL import Image

def preprocess(image_path):
input_shape = (1, 1, 64, 64)
img = Image.open(image_path)
img = img.resize((64, 64), Image.ANTIALIAS)
img_data = np.array(img)
img_data = np.resize(img_data, input_shape)
return img_data

Output

The model outputs a (1x8) array of scores corresponding to the 8 emotion classes, where the labels map as follows: emotion_table = {'neutral':0, 'happiness':1, 'surprise':2, 'sadness':3, 'anger':4, 'disgust':5, 'fear':6, 'contempt':7}

Postprocessing

Route the model output through a softmax function to map the aggregated activations across the network to probabilities across the 8 classes.

import numpy as np

def softmax(scores):
# your softmax function

def postprocess(scores):
'''
This function takes the scores generated by the network and returns the class IDs in decreasing
order of probability.
'''
prob = softmax(scores)
prob = np.squeeze(prob)
classes = np.argsort(prob)[::-1]
return classes

Sample test data

Sets of sample input and output files are provided in

• serialized protobuf TensorProtos (.pb), which are stored in the folders test_data_set_*/.

Quantization

Emotion FERPlus int8 is obtained by quantizing fp32 Emotion FERPlus model. We use Intel® Neural Compressor with onnxruntime backend to perform quantization. View the instructions to understand how to use Intel® Neural Compressor for quantization.

Prepare Model

Download model from ONNX Model Zoo.

wget https://github.com/onnx/models/raw/main/vision/body_analysis/emotion_ferplus/model/emotion-ferplus-8.onnx

Convert opset version to 12 for more quantization capability.

import onnx
from onnx import version_converter
model = onnx.load('emotion-ferplus-8.onnx')
model = version_converter.convert_version(model, 12)
onnx.save_model(model, 'emotion-ferplus-12.onnx')

Model quantize

cd neural-compressor/examples/onnxrt/body_analysis/onnx_model_zoo/emotion_ferplus/quantization/ptq_static
bash run_tuning.sh --input_model=path/to/model  \ # model path as *.onnx
--dataset_location=/path/to/data \
--output_model=path/to/save

License

MIT