Emphasise the optical flow is calculated at input resolution.

[5had3z]

Category:

Documentation

Description:

I personally have made a mistake in using this operator by misinterpreting the output vector quantities as the resolution of the output image. A common pattern pytorch to automate dealing with flow vectors is to autmatically scale the flow magnitude and resolution at the same time in the following code.

scale = image.shape[-1] / flow.shape[-1]
flow_scale = torch.nn.functional.interpolate(flow * scale, scale_factor=scale, mode='bilinear')

By emphasising the output of this operator is always at the resolution of the input, but the resolution of the output may differ, I hope to prevent others from making this same mistake.

Context

Original datapipe code...

pipedata["flow_pn"] = fn.optical_flow(image_seq)

...was fixed with (since default output_grid=4)

pipedata["flow_pn"] = fn.optical_flow(image_seq) / 4

Additional information:

Affected modules and functionalities:

Key points relevant for the review:

Tests:

• Existing tests apply
• New tests added
  • Python tests
  • GTests
  • Benchmark
  • Other
• N/A

Checklist

Documentation

• Existing documentation applies
• Documentation updated
  • Docstring
  • Doxygen
  • RST
  • Jupyter
  • Other
• N/A

DALI team only

Requirements

• Implements new requirements
• Affects existing requirements
• N/A

REQ IDs: N/A

JIRA TASK: N/A

[JanuszL]

!build

[JanuszL]

Hi @5had3z,

Thank you for your contribution.

[klecki]

!build

[dali-automaton]

CI MESSAGE: [13260491]: BUILD STARTED

[dali-automaton]

CI MESSAGE: [13260491]: BUILD PASSED