Hard, Yet Reducible: Controlled Forward Transfer for Synthetic Degradation Curation

Chunming He; Kailai Zhou; Jiaming Zuo; Hanqi Liu; Fengyang Xiao; Youwei Pang; Xiaofeng Liu; Weisi Lin

cs.CVOct 7, 2026

Hard, Yet Reducible: Controlled Forward Transfer for Synthetic Degradation Curation

Authors: Chunming He, Kailai Zhou, Jiaming Zuo, Hanqi Liu, Fengyang Xiao, Youwei Pang, Xiaofeng Liu, Weisi Lin, +1 more

Organizations: Tsinghua University, China · X3000 Inspection Co., Ltd · AI4X Team, Nanyang Technological University, Singapore · Yale University, USA

Abstract

Selecting synthetic degradations for dense prediction requires an estimate of their training utility, the generalization gain they bring under a finite training budget. Clean and degraded twins share content and labels, suggesting a score based on how much short training reduces the excess error caused by degradation. However, this gap can also shrink when clean performance deteriorates. Measuring the improvement on degraded images alone avoids that confound, but it still credits progress that the same amount of clean training would have produced. We propose the \textbf{controlled Reducible Degradation Gap} (cRDG) for regions defined by degradation type and severity. From a common checkpoint, cRDG runs two budget-matched probes that differ only in one augmentation slot, which holds either a synthetic degradation or a clean augmentation. The score is the gain on held-out degraded images relative to the clean-control probe. Clean harm is a separate feasibility constraint. cRDG reveals a correctable severity band in which training on the degradation yields high controlled gain under the available budget, and the band moves with the predictor, the starting checkpoint, and the training budget. \textbf{Curation of Reducible Bands} (\method) uses cRDG to select synthetic data without changing the predictor. On semantic segmentation and salient object detection, \method{} improves representative predictors under matched synthetic-data budgets and training schedules, extends to existing data-generation pipelines, and preserves clean performance. Code and supporting materials will be publicly released.

Figures & tables

Figure 1 : Why Matched Clean Control Matters. (a) Paired gap reduction can be inflated by clean forgetting. (b) Held-out degraded loss drop avoids this inflation but still credits generic training progress. (c) Our matched-control formulation underlying cRDG measures degraded gain $G_{g}$ beyond equally long clean training, while checking clean harm $H_{g}$ separately.

Figure 2 : Overview of CuRB . (a) Degradation-severity regions are generated and validity-filtered. (b) Two matched probes start from $\theta_{0}$ and differ only in the augmentation slot, where $\oplus$ denotes the union of the clean-replay stream and the slot stream. (c) Held-out evaluation yields degraded gain $G_{g}$ and clean harm $H_{g}$ . Averaging $G_{g}$ over the two split directions gives cRDG . $H_{g}$ determines feasibility, while the correctable band is diagnostic only. (d) Feasible positive-score regions are ranked and fill the budget $B_{\mathrm{sel}}$ in rank order, with $k$ -center sampling for the last partial region. (e) Final training restarts from $\theta_{0}$ on clean data, with $\mathcal{S}^{\star}$ in the synthetic slots under the fixed schedule.

Table 1 : Evaluation on SOD Benchmarks following the official protocol ( Section 4.1 ). At the same augmentation budget, comparing uncurated sampling with CuRB isolates the gain from curation.

		ACDC mIoU $\uparrow$		Dark Zurich mIoU $\uparrow$		Cityscapes
Data-generation host	Published host setting	Original	+ CuRB	Original	+ CuRB	val $\Delta$ $\uparrow$
single-generator hosts
ISSA [ 17 ]	SegFormer	52.45	53.92	27.39	28.76	$+0.24$
H-Weather [ 13 ]	SegFormer	49.21	50.63	23.44	24.65	$+0.36$
Gen4Seg [ 31 ]	SegFormer	51.04	52.39	25.63	27.28	$+0.45$
composite DG host

Table 2 : CuRB as a plug-in to existing data-generation settings. Published references for ISSA, H-Weather, and RobustNet+ISSA are taken from Li et al. [17] . Gen4Seg follows Yin et al. [31] (its Dark Zurich value is obtained with the released pipeline). Our “+ CuRB ” runs apply CuRB under the corresponding host setting and synthetic-data budget. The composite host stacks RobustNet [ 6 ] with ISSA on DeepLabV3+. Host-specific region construction: Appendix D .

	ACDC in-family			in-fam.	Snow	Dark	Cityscapes
Method	Fog	Night	Rain	mean	(held out)	Zurich	val
Source-only	60.61	28.42	50.35	46.46	48.79	24.08	67.84
matched curation protocol: same pool, gate, $B_{\mathrm{sel}}$ , objective, and schedule
All-gen. (uniform, $B_{\mathrm{sel}}$ )	63.82	31.51	52.78	49.37	50.15	26.55	68.21
Realism (NR-IQA / CLIPScore)	63.34	30.97	53.02	49.11	50.31	26.12	68.44
Highest loss / uncertainty	64.05	31.88	52.41	49.45	49.97	26.71	67.93

Table 3 : Controlled curation comparison on Cityscapes

{\to}

ACDC / Dark Zurich. All methods share the candidate pool, validity gate,

B_{\mathrm{sel}}

, final-training objective, and schedule, differing in their sample-selection policies. Fog, night, and rain are treated as the in-family conditions. Snow is held out as an unseen family, with no snow samples entering the candidate pool, curation, or final training.

Figure 3 : Qualitative comparison. (a) Degraded SOD with NUN. (b) adverse-condition SemSeg with ISSA SegFormer. Baseline and + CuRB predictions are shown for both tasks.

Table 7

Figure 4 : Training utility and the correctable band. (a) Per-bin retraining gains (top) and scores (bottom) for low light. Shading marks the predicted band. (b) Empirical severity centroids at three final budgets with scaled-

K

cRDG predictions overlaid (paired-bootstrap

95\%

CI). (c) cRDG centroids across probe budgets, checkpoints, and predictors. Error bars are s.d. over three partitions.

	Adapted valuation baselines			Twin-based scores
Metric	Highest loss / uncertainty	RHO-Loss style [ 19 ]	Grad. contrib. [ 34 ]	$\Delta_{0}$ (gap)	RDG (uncontrolled)	$D_{g}$ (held-out drop)	CuRB ( cRDG )
$\rho_{\mathrm{glob}}$ with $U_{\mathrm{real}}$ $\uparrow$	0.25	0.49	0.52	0.39	0.61	0.70	0.79
$\rho_{\mathrm{macro}}$ with $U_{\mathrm{real}}$ $\uparrow$	0.17	0.35	0.37	0.27	0.43	0.56	0.74
$\rho_{\mathrm{DZ}}$ with $U_{\mathrm{DZ}}$ $\uparrow$	0.21	0.43	0.45	0.33	0.53	0.62	0.72
in-fam. mIoU $\uparrow$	49.45	49.80	50.00	49.28	50.39	51.02	51.97

Table 6 : Region-level score comparison under matched pool, gate, diversity, budget, objective, and schedule (Cityscapes

{\to}

ACDC, ISSA SegFormer).

\rho_{\mathrm{glob}}

is Spearman correlation with

U_{\mathrm{real}}

over all

32

regions,

\rho_{\mathrm{macro}}

averages within-family correlations, and

\rho_{\mathrm{DZ}}

evaluates the same ranking against per-region retraining gains on Dark Zurich.

\Delta_{0}

is the unmastered gap and

D_{g}

the held-out loss drop.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Degradation	gate prec.	gate recall	accept. $s_{1}$ – $s_{4}$	accept. $s_{5}$ – $s_{8}$
Low light	0.94	0.91	0.99	0.95
Backlight	0.92	0.89	0.99	0.92
Fog	0.95	0.92	1.00	0.96
Rain	0.91	0.88	0.98	0.90

Appendix

Tab. S1 : Validity-gate audit on

500

sampled candidates per family. Structural drift is the positive class for gate precision and recall. Acceptance remains high at severe levels (

0.90

–

0.96

), indicating that the gate does not simply remove stronger degradations.

$\epsilon$	$0$	$0.5\,\mathrm{SE}$	$1\,\mathrm{SE}$ (default)	$2\,\mathrm{SE}$
feasible-region fraction	0.41	0.63	0.78	0.91
in-family mIoU $\uparrow$	51.43	51.86	51.97	51.79
Cityscapes mIoU $\uparrow$	70.21	70.24	70.13	69.82

Appendix

Tab. S2 : Sensitivity to the clean-retention tolerance

\epsilon

. Degraded-domain performance is stable around the default, while loosening the constraint permits a small clean-performance drop.

Quantity	Value
Regions $M$ (four families $\times$ eight severity bins)	$32$
Split directions $R$	$2$
Probe horizon $K$	$1/80$ of the final schedule
Treatment probes ( $MR$ )	$64$
Shared control probes ( $R$ )	$2$
Probe-step cost / final-step cost (measured)	$\approx 0.71$

Appendix

Tab. S3 : Curation cost. Operational selection uses one

A|B

partition. The region-independent control adds only two probes, and the normalized probe-update cost is

(66/80)\times 0.71\approx 0.59

of one final-training run, under the

K=T/80

scaling rule.

Explore similar work

Training-Free Metrics for Synthetic Object Detection Data: A Proxy for Detector Performance
Jun 18, 2026Myeongseok Nam, Donghun Yeo, Seungwook KimSynthetic Image Detection Synthetic Training Data
GenGenAI, South Korea
Post-Generation Curation of Synthetic Images via Homogeneous-Heterogeneous Splitting
Jul 2, 2026Disheng Liu, Tuo Liang, Chaoda Song +1Synthetic Image Detection Generative Models
Department of Computer and Data Sciences Case Western Reserve University Cleveland, OH, USA
Effective Synthetic Data Curation Requires Group-Level Signals
Sep 30, 2026Cathy Jiao, Chenyan XiongSynthetic Data Data-Curation
Language Technologies Institute, Carnegie Mellon University

Jun 18, 2026, cs.CVJ/K move · Enter open · S save

Training-Free Metrics for Synthetic Object Detection Data: A Proxy for Detector Performance

Myeongseok Nam, Donghun Yeo, Seungwook Kim

GenGenAI, South Korea

Synthetic Image Detection Synthetic Training Data Detection Performance Object Detection Training-Free

Open paper AskPDF arXiv

Synthetic images are increasingly used to augment scarce real data for object detection. However, not all synthetic sets help equally, and the only way to know a set's value is to train a detector on it, which is slow and demands dense annotation. We ask whether a training-free metric can instead rank candidate synthetic training sets by their downstream utility. Existing image-set metrics such as FID, KID, and MMD compare two feature distributions with a single global statistic, which we show is mis-specified for detection-data selection in two ways: it is blind to per-image composition (object count, box scale, class mix), and even at fixed composition its global averaging washes out the appearance differences that separate high-mAP pools from low-mAP ones. We propose Conditional-Composition Domain Match (CCDM), which converts any feature-space distance into a composition-stratified comparison, matching candidate and target within metadata-defined strata without training a detector. On COCO and VisDrone-DET, the best CCDM variant ranks 19 candidate training sets in strong agreement with YOLOv8 mAP (Spearman \r{ho} = 0.97 and 0.96), outperforming FID, KID, and MMD. Furthermore, CCDM holds when reference metadata comes from detector pseudo-labels rather than ground-truth boxes.

Abstract

Explore similar work

Training-Free Metrics for Synthetic Object Detection Data: A Proxy for Detector Performance

Post-Generation Curation of Synthetic Images via Homogeneous-Heterogeneous Splitting

Effective Synthetic Data Curation Requires Group-Level Signals