Switch branches/tags
Find file History
EmmaNingMS and ankkhedia Add demo for resnet50, squeezenet, mnist and emotion_ferplus (#118)
* Add demo link

* deleting the title by mistake

* Update the link

* Add the demo link

* Add demo link

* add demo link

* correct a mistake

* update the description

* Update README.md

* Update README.md

* Update README.md

* Update README.md

* Update README.md

* Update README.md
Latest commit 0bb8cc6 Dec 1, 2018
Permalink
Type Name Latest commit message Commit time
..
Failed to load latest commit information.
README.md Add demo for resnet50, squeezenet, mnist and emotion_ferplus (#118) Dec 1, 2018

README.md

FER+ Emotion Recognition

Description

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

Model

Model Checksum Download (with sample test data) ONNX version Opset version
Emotion FERPlus MD5 31 MB 1.0 2
MD5 31 MB 1.2 7
MD5 31 MB 1.3 8

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_*/.

License

MIT