FastDepth

Use case : Depth Estimation

Model description

FastDepth is a lightweight encoder-decoder network designed for real-time monocular depth estimation, optimized for edge devices. This implementation is based on model number 146 from PINTO's model zoo, which builds upon a MobileNetV1 based feature extractor and a fast decoder.

Although the original training dataset is not explicitly provided, it is most likely NYU Depth V2, a standard benchmark dataset for indoor depth estimation.

Network information

Network Information	Value
Framework	TensorFlowLite
Quantization	int8
Provenance	PINTO Model Zoo #146
Paper	Link to Paper

The models are quantized using tensorflow lite converter.

Network inputs / outputs

Input Shape	Description
(1, H, W, 3)	Single RGB image (int8)

Output Shape	Description
(1, H, W, 1)	Single-channel depth prediction (int8)

Recommended platforms

Platform	Supported	Recommended
STM32L0	[]	[]
STM32L4	[]	[]
STM32U5	[]	[]
STM32H7	[]	[]
STM32MP1	[]	[]
STM32MP2	[x]	[x]
STM32N6	[x]	[x]

Performances

Metrics

Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.

Reference NPU memory footprint

Model	Dataset	Format	Resolution	Series	Internal RAM (KiB)	External RAM (KiB)	Weights Flash (KiB)	STM32Cube.AI version	STEdgeAI Core version
Fast Depth	NYU depth v2	Int8	224x224x3	STM32N6	2365.98	0.0	1505.19	10.2.0	2.2.0
Fast Depth	NYU depth v2	Int8	256x256x3	STM32N6	2688	1024.0	1505.19	10.2.0	2.2.0
Fast Depth	NYU depth v2	Int8	224x224x3	STM32N6	2800	1600	1505.17	10.2.0	2.2.0

Reference NPU inference time

Model	Dataset	Format	Resolution	Board	Execution Engine	Inference time (ms)	Inf / sec	STM32Cube.AI version	STEdgeAI Core version
Fast Depth	NYU depth v2	Int8	224x224x3	STM32N6570-DK	NPU/MCU	24.43	40.93	10.2.0	2.2.0
Fast Depth	NYU depth v2	Int8	256x256x3	STM32N6570-DK	NPU/MCU	55.51	18.01	10.2.0	2.2.0
Fast Depth	NYU depth v2	Int8	320x320x3	STM32N6570-DK	NPU/MCU	56.07	17.83	10.2.0	2.2.0

Please refer to the stm32ai-modelzoo-services GitHub here