• Thermal Image Classification: Data Augmentation, Balancing, Time-Dependent Insights and Pretrained Models- Implemented a method to train neural networks with simulated thermal imaging from visible spectrum images. - Executed thorough data processing, enhancing model performance, and achieved a substantial reduction in classification errors, decreasing the error rate from 17.2% to 6.0%.• Learning to detect key points in a laboratory rat image in a video stream- Fine-tuning of the model was performed, which allowed to reduce RMSE for key points of the video stream by 5% and achieve an error rate of 2.9 pixels.Key technologies: Python, PyTorch, Numpy, DeepLabCut, Seaborn, Matplotlib, Pandas