Releases: haruishi43/py-edge-eval
Releases · haruishi43/py-edge-eval
Release Version 0.1.5
- non-IS Cityscapes evaluation for Post-SEAL
Version 0.1.4
- Fix bugs for
nms
- Compatible BSDS500 for blette
Version 0.1.3 Release
Improvements:
- Organize binary label evaluation codes
- Made it easier to create custom
Evaluator
Version 0.1.2 Release
Fix:
- support for PIL deprecations
- improve evaluator
Version 0.1.1 Release
Improvements:
- evaluation scripts
- organized imports
WIP:
- base/custom
Evaluators
Version 0.1.0 Release
Fixes:
mask2onehot
: input is nowlist
of ids instead ofnum_classes
; this enables more flexibility and robustnessmask2edge
: ignore labels are not indices
Version 0.0.5 Release
Features:
- Introduce
Evaluator
class for dataset specific evaluation - Edge generation tools (
Mask2Edge
andmask2edge
) - Clearer evaluation parameters (Pre-SEAL, Post-SEAL and Raw and Thin)
- NMS inside evaluation
- Half and OTF evaluation for Cityscapes (along with GT edge generation script)
- Multiprocessing
mask2edge
Fixes:
- ODS metrics were not interpolated. This was fixed and made sure that the logic is true to the original MATLAB evaluation script.
Refactor:
- Directory organization
- Remove
partial
functions
Version 0.0.4 Release
What's new?
- Full support for semantic boundary detection tasks
- Full supports for BSDS500, SBD, and Cityscapes
- Full support for evaluation protocols such as "raw", "thin" and instance-sensitive edges
- Tools for converting Cityscapes datasets
- Modules for generating PR-Curves
- Checks for valid results (compared results from MATLAB)
SBD evaluation codes and multiprocessing
Added:
- Faster evaluation using multiprocessing
- Evaluation code for SBD
- Benchmark codes (test) for BSDS500 and SBD
Development Version [0.0.3-dev0]
The main functionalities are there:
- pixel matching
- NMS
- supporting functions
- example scripts
- test codes
For the upcoming release, I need to add:
- Better BSDS evaluation script with multiprocessing (currently takes around 3 hours for evaluation)
- Support for semantic boundary datasets such as SBD