Remove eoscale

[namsellem]

Objective

Remove the dependency on Eoscale from the processing pipeline in order to:

• Implement eomultiprocessing logic independently from shared memory logic
• Prepare future refactoring and modularization in compliance with Bulldozer
• Validate Performances CPU/RAM usage along the workflow

---

Scope

This issue tracks the global removal of Eoscale from the main processing components.
Each component removal may be handled in dedicated sub-issues for better traceability.

---

Tasks

Remove Eoscale from the following components (to be broken down into dedicated issues if needed):

• Prepare

  • Identify Eoscale usage
  • Refactor logic to remove dependency
  • Validate outputs consistency
  • Update unit/integration tests if needed
  • Validate non-regression through automated tests
    eoscale removal in prepare #47

• Watermask

  • Identify Eoscale usage
  • Refactor implementation
  • Validate outputs consistency
  • Update unit/integration tests if needed
  • Validate non-regression through automated tests:
    eoscale removal in watermask #48

• Shadowmask

  • Identify Eoscale usage
  • Refactor implementation
  • Validate outputs consistency
  • Update unit/integration tests if needed
  • Validate non-regression through automated tests
    Remove eoscale shadow stack #63

• Vegetationmask

  • Identify Eoscale usage
  • Refactor implementation
  • Validate outputs consistency
  • Update unit/integration tests if needed
  • Validate non-regression through automated tests

• Urbanmask

  • Identify Eoscale usage
  • Refactor implementation
  • Validate outputs consistency
  • Update unit/integration tests if needed
  • Validate non-regression through automated tests
    eoscale revoval in urbanmask #49

• Stackmask

  • Identify Eoscale usage
  • Refactor implementation
  • Validate outputs consistency
  • Update unit/integration tests if needed
  • Validate non-regression through automated tests
    Remove eoscale shadow stack #63

---

Validation Criteria

• No remaining Eoscale imports in the codebase
• Pipeline runs successfully end-to-end
• No regression in outputs
• All tests passing
• Documentation updated if required

---

Notes

This issue acts as a tracking issue.
Sub-issues will be linked here for detailed implementation tracking.

[ytanguy]

@namsellem : I've checked watermask and some info outputs are missing :

• [3] Select samples
• [4] Learn model
• [5] Predict
• [6] Post-process

It will help follow the algorithm execution. The same applies to other algorithm (like urbanmask).

[namsellem]

Performace Overview

Here's a performance overview of slurp's "remove_eoscale" branch before merging #69 on a (10000*14000) sample image.
The tile_max_size argument is set to 1000, 8 CPUs were used for this test, 140 threads per mp_n_to_m_images were executed.