v0.4.0

spandrel v0.4.0 / spandrel-extra-arches v0.2.0

What's Changed

• Add detection for PLKSR's lk_type parameter by @RunDevelopment in #264
• Fix size requirements for 1x AdaCode models by @RunDevelopment in #267
• Add support for RetinexFormer by @RunDevelopment in #270
• Add support for HVI-CIDNet by @RunDevelopment in #271
• Fix NumPy v2 incompatible with OpenCV in tests by @RunDevelopment in #278
• Restore training conditionals for PLKSR, RealPLKSR, SPAN by @the-database in #276
• Require keyword args for all architectures by @RunDevelopment in #279
• Rename all arch directories to __arch by @RunDevelopment in #283
• Fixed scripts using installed versions of spandrel by @RunDevelopment in #286
• Improve initialization for extra arches by @akx in #281
• Fixed input channels and size req of LaMa by @RunDevelopment in #289
• Include source distributions by @RunDevelopment in #290
• Ensure that exported architecture class names match the arch ID/name by @RunDevelopment in #284
• Make CUGAN input/output channels configurable for all versions by @RunDevelopment in #285
• Define the public API by what's documented by @RunDevelopment in #287
• Fixed that MAT cannot load its own models by @RunDevelopment in #291
• Add support for SeemoRe by @RunDevelopment in #294
• Add support for RealPLKSR DySample by @joeyballentine in #293
• Fix RPLKSR DySample by @joeyballentine in #297
• Add support for MoSR by @the-database in #307

New Contributors

• @the-database made their first contribution in #276

Full Changelog: v0.3.4...v0.4.0

v0.3.4

What's Changed

• Optimize call api by only calling .eval if in training mode by @joeyballentine in #261

Full Changelog: v0.3.3...v0.3.4

v0.3.3

What's Changed

• Make architecture fields Final by @RunDevelopment in #254

New architectures

• PLKSR & RealPLKSR support by @joeyballentine in #252

Full Changelog: v0.3.2...v0.3.3

v0.3.2

This is the release for spandrel==0.3.2.

What's Changed

• Error-tolerant loading for .pt files by @RunDevelopment in #215
• Removed unused code in GRL and HAT by @RunDevelopment in #217 and #226
• Fix call API for RestoreFormer by @RunDevelopment in #219
• Make ATD compatible with neosr by @RunDevelopment in #240
• Minor improvements to multiple architectures by @RunDevelopment in #218, #244, #249
• Numerous improvements to tests and our CI setup in #220, #221, #222, #225, #241, #242, #243
• Update to ruff 0.4.0 by @joeyballentine in #239
• Add tensor shape reference by @RunDevelopment in #216

New architectures

• Add support for IPT by @RunDevelopment in #223
• Add support for SAFMN_BCIE by @RunDevelopment in #231
• Add support for DRCT by @RunDevelopment in #248

Full Changelog: v0.3.1...v0.3.2

v0.3.1

This is the release for both spandrel==0.3.1 and spandrel_extra_arches==0.1.1.

What's Changed

• Support APISR models (and others) by improving how state dicts are interpreted by @RunDevelopment in #205 and #206
• Slim down FeMaSR for inference by @RunDevelopment in #207

New architectures

• Add support for MixDehazeNet by @RunDevelopment in #204
• Add support for ATD by @RunDevelopment in #200
• Add support for AdaCode by @RunDevelopment in #208
• Add support for MPRNet by @RunDevelopment in #210
• Add support for MIRNet2 by @RunDevelopment in #212
• Add support for DnCNN, FDnCNN, and DRUNet by @RunDevelopment in #214

Full Changelog: v0.3.0...v0.3.1

v0.3.0

This is the release for both spandrel==0.3.0 and spandrel_extra_arches==0.1.0.

What's Changed

• Change project license to MIT license by @RunDevelopment in #156
  All code in spandrel, including architecture code, is now licensed under permissive licenses. Architectures with restrictive licenses have been moved to the new spandrel_extra_arches package.
• Optimize for inference when using call api by @joeyballentine in #162
• Add internal hyperparameter API by @RunDevelopment in #164
• Support UpCunet2x_Fast for Real-CUGAN by @RunDevelopment in #188
• Improved DAT detection by @RunDevelopment in #167
• Make it possible to save SPAN models as .safetensors by @RunDevelopment in #157
• Remove SPSR due to wrong license by @RunDevelopment in #175
• Remove unused Real-ESRGAN code by @RunDevelopment in #176
• Make ArchRegistry iterable by @RunDevelopment in #174

New architectures

• Add support for RGT by @RunDevelopment in #168
• Add support for Restormer by @RunDevelopment in #170
• Add support for FFTformer by @RunDevelopment in #171
• Add support for M3SNet by @RunDevelopment in #172
• Add support for DCTLSA by @RunDevelopment in #147

Full Changelog: v0.2.2...v0.3.0

v0.2.2

What's Changed

• Allow device strings in to again by @RunDevelopment in #139
• Fix Google Drive downloads by @RunDevelopment in #140
• Fixed type error due to PyTorch 2.2 by @RunDevelopment in #149
• Add SPAN norm parameter by @RunDevelopment in #148

New architectures

• Add support for NAFNet by @RunDevelopment in #141

Full Changelog: v0.2.1...v0.2.2

v0.2.1

What's Changed

• Pad images to size requirements in call API by @RunDevelopment in #137

Full Changelog: v0.2.0...v0.2.1

v0.2.0

What's Changed

• Breaking: Rename GRLIR to GRL by @RunDevelopment in #120
• Breaking: More guarantees for minimum and multiple_of in SizeRequirements by @RunDevelopment in #124
• Breaking: Bring model.to signature in line with PyTorch module.to by @RunDevelopment in #133
• Breaking: Verify dtype in to by @RunDevelopment in #135
• Save Real-CUGAN models correctly by @RunDevelopment in #119
• Fix ESRGAN size requirement by @RunDevelopment in #121
• Clamp output tensor to guarantee output range by @RunDevelopment in #127
• Deduplicate tensor pad-to-size code by @akx in #104
• Add dtype, cpu, and cuda to model API by @RunDevelopment in #132

New architectures

• Add support for SAFMN by @RunDevelopment in #130

Full Changelog: v0.1.8...v0.2.0

v0.1.8

What's Changed

• Allow strings for Torch devices, like Torch does by @akx in #112
• Add device property for models by @RunDevelopment in #114
• Removed unnecessary CUGAN code by @RunDevelopment in #117
• arch: remove torch.load calls by @akx in #109
• Add "Used By" section to README by @joeyballentine in #102
• Lint with pre-commit by @akx in #94
• Simplify scripts by @akx in #108
• tests: avoid sys.path hacks by @akx in #106
• tests: force snapshots to have LF line-endings by @akx in #110
• tests: add device and outputs directory environment variables by @akx in #111

Full Changelog: v0.1.7...v0.1.8