ELDepth

ELDepth: Extreme Low-Power Monocular Depth Estimation for mobile devices

Models evaluated

• U-Net: Convolutional Networks for Biomedical Image Segmentation
• Squeeze U-Net: A Memory and Energy Efficient Image Segmentation Network
• LiteDepth: Digging into Fast and AccurateDepth Estimation on Mobile Devices

Datasets

• DIODE: A Dense Indoor and Outdoor DEpth Dataset
• NYU Depth Dataset V2
• The KITTI Vision Benchmark Suite

Future work

• Model destilltion (base code)[https://keras.io/examples/keras_recipes/better_knowledge_distillation/]
• Dynamic quantization
• SSL
• visualization as in https://github.com/aim-uofa/AdelaiDepth