garryling/depth_anything_v2_optimized

🔍 Depth Anything V2 - Ultra Fast

State-of-the-Art Depth Estimation Meets Mobile Efficiency
Optimized for Snapdragon NPUs with 2X Speed Boost

⚡ Benchmark Comparison

• Input resolution:518x518 Snapdragon 8 Gen2 (NPU Acceleration)

  Metric	depth_anything_v2.onnx	depth_anything_v2_mha.onnx	Improvement
  Latency	152ms	73ms	2.08×
  Memory	102MB	45MB	2.26×

benchmark method

please intall qai_hub first

import qai_hub as hub

compile_job = hub.submit_compile_job(
    model="depth_anything_v2.onnx",
    device=hub.Device("Samsung Galaxy S23 (Family)"),
)
assert isinstance(compile_job, hub.CompileJob)

profile_job = hub.submit_profile_job(
    model=compile_job.get_target_model(),
    device=hub.Device("Samsung Galaxy S23 (Family)"),
)

reference

• qaihub
• qai depth anything result