SoyabulIslamLincoln/EmotionsInTheWild-CNN-Benchmarks

Emotion (Context + Facial) recognition in the wild using ConvNets (EfficientNet, ResNet, ResNext)

Emotion (Context + Facial) recognition in the wild using ConvNets (EfficientNet, ResNet, ResNext)

---

Emotions In the Wild: CNN benchmarks

Emotion (Context + Facial) recognition in the wild using ConvNets (EfficientNet, ResNet, ResNext)

For emotic dataset (pre-training),

• Download the dataset from http://sunai.uoc.edu/emotic/download.html

• Download the json annotations from https://www.kaggle.com/furcifer/emoticlabelsjson

• Install dependencies (python >= 3.5)

  pip install -r requirements.txt

• If you want to use docker,

  nvidia-docker build -t nabil/efncv:eitw .
  nvidia-docker run -it -d -v /path_to_github_repo/EmotionsInTheWild-CNN-Benchmarks/:/eitw/ --net=host --ipc=host nabil/efncv:eitw /bin/bash
  

Training

• Set the hyperparameters in the config.py file

• Run, train.py

Pre-trained models

Model	Dataset	pre-trained weight
resnet34	emotic	resnet34
resnet50	emotic + cfid	resnet50
efficientnet-b0	emotic + cfid	efn-b0
efficientnet-b1	emotic + cfid	efn-b1
efficientnet-b2	emotic + cfid	efn-b2

Support

Tested with: python3.6 python3.7 python3.8 Model support: resnet 18 to 152 resnext 50, 101 efficientnets b0 to b7

TO-DO:

• multi-gpu training
• augmentation
• encoding facial landmarks

Others

Any contribution is welcome.

• Contributors