This repository is in association to a research paper submitted to the Animal-Computer Interaction Conference, scheduled on December 2nd, in Glasgow, UK.
We have experimented with three different contrastive learning frameworks, belonging to two different contrastive learning appraches, namely Simple Contrastive Learning and Momentum Contrast.
These models were trained on a private original that we assembled over a period of 3 months. The seven labels are directly inspired from the Panksepp Seven Emotions.
| Emotion | Supervised ResNet-50 Accuracy | Unsupervised MoCo-v1 Accuracy |
|---|---|---|
| Caring | 94.74% | 34.61% |
| Exploring | 83.75% | 40.40% |
| Fear | 28.95% | 35.51% |
| Lust | 47.05% | 62.79% |
| Playing | 46.34% | 38.88% |
| Rage | 87.09% | 45.91% |
| Sadness | 78.57% | 44.37% |
Access our best model with a modified ResNet-34 encoder.
wget https://raw.githubusercontent.com/caffeinekeyboard/Dog_Emotion_Classification/master/MoCo_Trials/labeled_Aarya/kaggle_sessions/session_APR11_953_96_R34/example_saved_models/encQ_best.pthHere are some related projects