Each grid shows 32 samples per class, fixed noise seed 101. First four rows: cat. Last four rows: dog. No training photographs are embedded.
Inspect structure, variety and class agreement. Losses and pixel diversity do not establish semantic quality. Zero duplicate pairs can coexist with poor images. A short single-seed pilot cannot validate a pet-image recipe or quantum advantage.
Manual inspection of these short trials still shows blurry textures and blobs; no convincing pet-image improvement is demonstrated.
1000 updates; current sampling weights.
1000 updates; EMA sampling weights.
1000 updates; current sampling weights.
1000 updates; EMA sampling weights.
1000 updates; EMA sampling weights.
{
"software": {
"torch": "2.14.1+cpu",
"numpy": "2.5.4",
"python": "3.12.14 (main, Aug 25 2026, 14:00:49) [Clang 22.1.3 ]",
"platform": "Linux-6.18.55-x86_64-with-glibc2.39"
},
"device": "cpu",
"threads": 1,
"source": {
"kind": "UofT CIFAR-10 train mirror",
"url": "https://huggingface.co/datasets/uoft-cs/cifar10/resolve/main/plain_text/train-00000-of-00001.parquet",
"file_sha256": "8428b53a88a11ac374111006708df51469e315a22ac6d66470afd9c78d2ae883"
},
"seed": 7,
"batch_size": 32,
"latent_dim": 16,
"classes": [
"cat",
"dog"
],
"training_samples": 1600,
"validation_samples": 400,
"training_fingerprint": "d52a678539f87d0ed763c37eed311212cccfe0c34faddd17417259407fede23f",
"validation_fingerprint": "0a375ae67d972c72b48769ff31e5d149d686372de7e6e2f579f6b9826db5cda1",
"image_shape": [
3,
32,
32
],
"evaluation_seed": 101,
"evaluation_samples_per_class": 32,
"semantic_quality": "not automatically evaluated",
"comparison": "Same data/order/prior/updates; model parameter counts differ. Regularized hybrid changes readout, angle map and bypass together, not an ablation.",
"limitations": [
"Single seed and bounded steps; not validation of the 64x64 Kaggle pet workflow.",
"Training times are single-session records, not comparative performance benchmarks.",
"Pixel diversity is not semantic quality; no quantum advantage claim."
],
"models": [
{
"case": "classical-legacy",
"generator_config": {
"kind": "classical",
"image_size": 32,
"channels": 3,
"latent_dim": 16,
"num_classes": 2,
"hidden": 64,
"seed": 7,
"architecture": "conv"
},
"trainer_config": {
"learning_rate": 0.0002,
"discriminator_hidden": 64,
"seed": 7,
"device": "cpu",
"discriminator_architecture": "conv",
"horizontal_flip": true,
"training": "legacy",
"generator_learning_rate": 0.0002,
"discriminator_learning_rate": 0.0002,
"r1_gamma": 0.0,
"r1_every": 16,
"diff_augment": false,
"ema_decay": 0.0
},
"steps": 1000,
"generator_parameters": 45491,
"discriminator_parameters": 82353,
"sampling_weights": "current",
"training_seconds": 54.84952731599992,
"timing_scope": "Training and logging, synchronized CUDA; excludes setup, sampling and export. Single session, no repeated timing benchmark.",
"before": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.047240838408470154,
"generated_pixel_std": 0.0019719202537089586,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.0022549026180058718,
"closest_pair_pixel_mae": 0.001300041563808918,
"near_duplicate_pair_fraction": 1.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.0017376767937093973
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.04934289678931236,
"generated_pixel_std": 0.0019741745200008154,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.002251187339425087,
"closest_pair_pixel_mae": 0.0013288859045132995,
"near_duplicate_pair_fraction": 1.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.0017339212354272604
}
},
"after": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.06526779383420944,
"generated_pixel_std": 0.21235567331314087,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.24541303515434265,
"closest_pair_pixel_mae": 0.11145149916410446,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.21686206758022308
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.06739222258329391,
"generated_pixel_std": 0.19685523211956024,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.22619205713272095,
"closest_pair_pixel_mae": 0.1276768445968628,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.19636711478233337
}
},
"raw": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.06526779383420944,
"generated_pixel_std": 0.21235567331314087,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.24541303515434265,
"closest_pair_pixel_mae": 0.11145149916410446,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.21686206758022308
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.06739222258329391,
"generated_pixel_std": 0.19685523211956024,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.22619205713272095,
"closest_pair_pixel_mae": 0.1276768445968628,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.19636711478233337
}
},
"final_losses": {
"discriminator_loss": 0.684712290763855,
"generator_loss": 0.6856949925422668,
"generated_mean": 0.455648809671402,
"step": 1000,
"epoch": 19
},
"preview_path": "classical-legacy/after.png"
},
{
"case": "classical-regularized",
"generator_config": {
"kind": "classical",
"image_size": 32,
"channels": 3,
"latent_dim": 16,
"num_classes": 2,
"hidden": 64,
"seed": 7,
"architecture": "conv"
},
"trainer_config": {
"learning_rate": 0.0002,
"discriminator_hidden": 64,
"seed": 7,
"device": "cpu",
"discriminator_architecture": "conv",
"horizontal_flip": true,
"training": "regularized",
"generator_learning_rate": 0.0002,
"discriminator_learning_rate": 0.0001,
"r1_gamma": 1.0,
"r1_every": 16,
"diff_augment": true,
"ema_decay": 0.995
},
"steps": 1000,
"generator_parameters": 45491,
"discriminator_parameters": 82353,
"sampling_weights": "EMA",
"training_seconds": 47.531684854000105,
"timing_scope": "Training and logging, synchronized CUDA; excludes setup, sampling and export. Single session, no repeated timing benchmark.",
"before": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.047240838408470154,
"generated_pixel_std": 0.0019719202537089586,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.0022549026180058718,
"closest_pair_pixel_mae": 0.001300041563808918,
"near_duplicate_pair_fraction": 1.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.0017376767937093973
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.04934289678931236,
"generated_pixel_std": 0.0019741745200008154,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.002251187339425087,
"closest_pair_pixel_mae": 0.0013288859045132995,
"near_duplicate_pair_fraction": 1.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.0017339212354272604
}
},
"after": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.09823013097047806,
"generated_pixel_std": 0.2364615797996521,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.2765565514564514,
"closest_pair_pixel_mae": 0.1101156696677208,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.25646641850471497
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.07333892583847046,
"generated_pixel_std": 0.22451017796993256,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.26196086406707764,
"closest_pair_pixel_mae": 0.11475091427564621,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.24120013415813446
}
},
"raw": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.08875477313995361,
"generated_pixel_std": 0.1876625269651413,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.217312291264534,
"closest_pair_pixel_mae": 0.12957139313220978,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.1857052892446518
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.10775163024663925,
"generated_pixel_std": 0.17141741514205933,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.19709688425064087,
"closest_pair_pixel_mae": 0.11608944088220596,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.1650477647781372
}
},
"final_losses": {
"discriminator_loss": 0.687303364276886,
"generator_loss": 0.7050588130950928,
"r1_penalty": 0.0,
"discriminator_total_loss": 0.687303364276886,
"real_probability": 0.499570369720459,
"fake_probability": 0.49326103925704956,
"generated_mean": 0.44602271914482117,
"step": 1000,
"epoch": 19
},
"preview_path": "classical-regularized/after.png"
},
{
"case": "hybrid-legacy",
"generator_config": {
"kind": "hybrid",
"image_size": 32,
"channels": 3,
"latent_dim": 16,
"num_classes": 2,
"n_qubits": 4,
"depth": 1,
"hidden": 64,
"seed": 7,
"architecture": "conv",
"readout": "z",
"angle_map": "tanh",
"latent_skip": false
},
"trainer_config": {
"learning_rate": 0.0002,
"discriminator_hidden": 64,
"seed": 7,
"device": "cpu",
"discriminator_architecture": "conv",
"horizontal_flip": true,
"training": "legacy",
"generator_learning_rate": 0.0002,
"discriminator_learning_rate": 0.0002,
"r1_gamma": 0.0,
"r1_every": 16,
"diff_augment": false,
"ema_decay": 0.0
},
"steps": 1000,
"generator_parameters": 31243,
"discriminator_parameters": 82353,
"sampling_weights": "current",
"training_seconds": 62.14072174099988,
"timing_scope": "Training and logging, synchronized CUDA; excludes setup, sampling and export. Single session, no repeated timing benchmark.",
"before": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.045005083084106445,
"generated_pixel_std": 0.0006110862013883889,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.0006763377459719777,
"closest_pair_pixel_mae": 2.286352355440613e-05,
"near_duplicate_pair_fraction": 1.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.0005269299144856632
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.04737091436982155,
"generated_pixel_std": 0.0007969803991727531,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.0008953406359069049,
"closest_pair_pixel_mae": 6.219524220796302e-05,
"near_duplicate_pair_fraction": 1.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.000713532674126327
}
},
"after": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.14580214023590088,
"generated_pixel_std": 0.13682901859283447,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.15178093314170837,
"closest_pair_pixel_mae": 0.016988394781947136,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.14473384618759155
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.1351657211780548,
"generated_pixel_std": 0.16220897436141968,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.1864340752363205,
"closest_pair_pixel_mae": 0.009030429646372795,
"near_duplicate_pair_fraction": 0.002016128972172737,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.17723290622234344
}
},
"raw": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.14580214023590088,
"generated_pixel_std": 0.13682901859283447,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.15178093314170837,
"closest_pair_pixel_mae": 0.016988394781947136,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.14473384618759155
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.1351657211780548,
"generated_pixel_std": 0.16220897436141968,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.1864340752363205,
"closest_pair_pixel_mae": 0.009030429646372795,
"near_duplicate_pair_fraction": 0.002016128972172737,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.17723290622234344
}
},
"final_losses": {
"discriminator_loss": 0.7503820657730103,
"generator_loss": 0.7662668228149414,
"generated_mean": 0.4554048478603363,
"step": 1000,
"epoch": 19
},
"preview_path": "hybrid-legacy/after.png",
"hybrid_bottleneck": {
"readout": "z",
"angle_map": "tanh",
"latent_skip": false,
"quantum_feature_dimension": 4,
"decoder_inputs": "quantum features only",
"classes": {
"0": {
"sample_count": 32,
"mean_angle_std": 1.3189984560012817,
"mean_quantum_feature_std": 0.26355981826782227,
"tanh_saturation_fraction": 0.0
},
"1": {
"sample_count": 32,
"mean_angle_std": 0.9376826286315918,
"mean_quantum_feature_std": 0.4177634119987488,
"tanh_saturation_fraction": 0.2109375
}
}
}
},
{
"case": "hybrid-regularized",
"generator_config": {
"kind": "hybrid",
"image_size": 32,
"channels": 3,
"latent_dim": 16,
"num_classes": 2,
"n_qubits": 4,
"depth": 1,
"hidden": 64,
"seed": 7,
"architecture": "conv",
"readout": "probabilities",
"angle_map": "linear",
"latent_skip": true
},
"trainer_config": {
"learning_rate": 0.0002,
"discriminator_hidden": 64,
"seed": 7,
"device": "cpu",
"discriminator_architecture": "conv",
"horizontal_flip": true,
"training": "regularized",
"generator_learning_rate": 0.0002,
"discriminator_learning_rate": 0.0001,
"r1_gamma": 1.0,
"r1_every": 16,
"diff_augment": true,
"ema_decay": 0.995
},
"steps": 1000,
"generator_parameters": 59915,
"discriminator_parameters": 82353,
"sampling_weights": "EMA",
"training_seconds": 53.27030629900014,
"timing_scope": "Training and logging, synchronized CUDA; excludes setup, sampling and export. Single session, no repeated timing benchmark.",
"before": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.0540425069630146,
"generated_pixel_std": 0.0013030931586399674,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.0014728870010003448,
"closest_pair_pixel_mae": 0.0008635413832962513,
"near_duplicate_pair_fraction": 1.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.0011364505626261234
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.05673213675618172,
"generated_pixel_std": 0.0012634285958483815,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.00144716736394912,
"closest_pair_pixel_mae": 0.0008652757969684899,
"near_duplicate_pair_fraction": 1.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.0010890534613281488
}
},
"after": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.15506483614444733,
"generated_pixel_std": 0.23530615866184235,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.263053834438324,
"closest_pair_pixel_mae": 0.08546694368124008,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.2476530522108078
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.11313825100660324,
"generated_pixel_std": 0.25599828362464905,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.2912386655807495,
"closest_pair_pixel_mae": 0.13550764322280884,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.2704007625579834
}
},
"raw": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.06875661015510559,
"generated_pixel_std": 0.18000106513500214,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.209034264087677,
"closest_pair_pixel_mae": 0.10252238065004349,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.19059181213378906
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.08573292940855026,
"generated_pixel_std": 0.1896224021911621,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.219624325633049,
"closest_pair_pixel_mae": 0.12224629521369934,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.1980389952659607
}
},
"final_losses": {
"discriminator_loss": 0.691534161567688,
"generator_loss": 0.6918132305145264,
"r1_penalty": 0.0,
"discriminator_total_loss": 0.691534161567688,
"real_probability": 0.5028631091117859,
"fake_probability": 0.500719428062439,
"generated_mean": 0.4461403787136078,
"step": 1000,
"epoch": 19
},
"preview_path": "hybrid-regularized/after.png",
"hybrid_bottleneck": {
"readout": "probabilities",
"angle_map": "linear",
"latent_skip": true,
"quantum_feature_dimension": 16,
"decoder_inputs": "quantum features plus classical noise",
"classes": {
"0": {
"sample_count": 32,
"mean_angle_std": 0.4072578549385071,
"mean_quantum_feature_std": 0.025567755103111267,
"tanh_saturation_fraction": null
},
"1": {
"sample_count": 32,
"mean_angle_std": 0.43876758217811584,
"mean_quantum_feature_std": 0.04177941009402275,
"tanh_saturation_fraction": null
}
}
}
},
{
"case": "classical-dcgan",
"generator_config": {
"kind": "classical",
"image_size": 32,
"channels": 3,
"latent_dim": 16,
"num_classes": 2,
"hidden": 64,
"seed": 7,
"architecture": "dcgan"
},
"trainer_config": {
"learning_rate": 0.0002,
"discriminator_hidden": 64,
"seed": 7,
"device": "cpu",
"discriminator_architecture": "dcgan",
"horizontal_flip": true,
"training": "regularized",
"generator_learning_rate": 0.0002,
"discriminator_learning_rate": 0.0001,
"r1_gamma": 1.0,
"r1_every": 16,
"diff_augment": true,
"ema_decay": 0.995
},
"steps": 1000,
"generator_parameters": 733059,
"discriminator_parameters": 673217,
"sampling_weights": "EMA",
"training_seconds": 168.8707305979999,
"timing_scope": "Training and logging, synchronized CUDA; excludes setup, sampling and export. Single session, no repeated timing benchmark.",
"before": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.049390364438295364,
"generated_pixel_std": 0.0006328251329250634,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.000727510021533817,
"closest_pair_pixel_mae": 0.0004255290550645441,
"near_duplicate_pair_fraction": 1.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.00018353147606831044
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.051965463906526566,
"generated_pixel_std": 0.0006468408391810954,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.000745846948120743,
"closest_pair_pixel_mae": 0.00048451623297296464,
"near_duplicate_pair_fraction": 1.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.0001904229720821604
}
},
"after": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.08027297258377075,
"generated_pixel_std": 0.18741029500961304,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.21761146187782288,
"closest_pair_pixel_mae": 0.060181017965078354,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.21060484647750854
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.066044382750988,
"generated_pixel_std": 0.1848437786102295,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.21426346898078918,
"closest_pair_pixel_mae": 0.07570017129182816,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.20747460424900055
}
},
"raw": {
"cat": {
"sample_count": 32,
"class_mean_pixel_mae": 0.09211916476488113,
"generated_pixel_std": 0.23399828374385834,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.26755964756011963,
"closest_pair_pixel_mae": 0.08604830503463745,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.25825780630111694
},
"dog": {
"sample_count": 32,
"class_mean_pixel_mae": 0.07396414875984192,
"generated_pixel_std": 0.2405286580324173,
"pair_count": 496,
"mean_pairwise_pixel_mae": 0.27784594893455505,
"closest_pair_pixel_mae": 0.08524558693170547,
"near_duplicate_pair_fraction": 0.0,
"near_duplicate_threshold": 0.01,
"mean_pairwise_pooled_mae": 0.26791343092918396
}
},
"final_losses": {
"discriminator_loss": 0.7129061818122864,
"generator_loss": 0.6782514452934265,
"r1_penalty": 0.0,
"discriminator_total_loss": 0.7129061818122864,
"real_probability": 0.49074888229370117,
"fake_probability": 0.5058045387268066,
"generated_mean": 0.4417050778865814,
"step": 1000,
"epoch": 19
},
"preview_path": "classical-dcgan/after.png"
}
],
"observed_quality": "Manual inspection of these short trials still shows blurry textures and blobs; no convincing pet-image improvement is demonstrated.",
"reproduction": "python benchmarks/image_stability.py --cifar-parquet FILE --steps 1000 --output .release-artifacts/stability-real"
}