Emotion Recognition with Mobile-Optimised Deep Learning

Problem

State-of-the-art emotion recognition models were too large and power-hungry for on-device mobile deployment. Running inference in the cloud introduced latency, privacy concerns, and dependency on connectivity.

Approach

Designed a multi-phase pipeline: model selection and benchmarking on FER-2013, fine-tuning on RAF-DB for improved real-world accuracy, then dynamic INT8 quantisation to shrink the model footprint for mobile CPUs. Built the pipeline with PyTorch and OpenCV.

Result

Runs offline on clinical devices, no cloud, no latency, no privacy risk. Patients get emotion detection without their data leaving the room.

• ◆Runs offline on clinical devices with no cloud dependency
• ◆Patient data never leaves the device, full privacy
• ◆4x smaller model with competitive accuracy retained
• ◆Real-time inference without network latency