acceptodds
Desk rejected at ICLR 2027

PAVER: Preserving Across-scale Visual Evidence with Risk-conditioned Interface for Remote-Sensing Tiny-Object Detection

Desk rejected at ICLR 2027, so left out of the paper lists and the map.

Abstract

Remote-sensing detectors must recover few-pixel targets from cluttered wide-area imagery and assign informative confidence to each candidate. Both demands share an evidence-retention bottleneck: weak cues must survive feature formation and cross-scale propagation, then remain accessible at inference. Preserving Across-scale Visual Evidence with Risk-conditioned Interface (PAVER) is a framework combining an evidence-preserving detector with a separately fitted Risk-conditioned Interface. The Complementary State-Flow Mixer (CSFM), Semantic–Contour Recalibration Gate (SCRG), and High-Resolution Pyramid Extension (HRPE) maintain complementary, localization-sensitive, and high-resolution pathways. The interface makes candidate confidence more trustworthy by recalibrating each fixed detection toward its likelihood of being a true positive, using a false-positive–true-positive prototype margin from P2 and P3 ROI evidence without changing boxes or classes. Under controlled reproduction, PAVER improves YOLOv12-s by 5.7 and 4.0 points in and , respectively, on VisDrone, with complementary evaluation on RSOD and DIOR. On VisDrone test-dev, risk conditioning gives slightly favorable calibration point estimates over Platt scaling while keeping essentially unchanged; all paired 95% CIs include zero, and temperature scaling retains the best ECE and NLL. PAVER combines strong detection with candidate-local confidence conditioning through its optional Risk-conditioned Interface. Code is publicly available at [https://anonymous.4open.science/r/PAVER-F47C/](https://anonymous.4open.science/r/PAVER-F47C/).

Primary area:
applications to computer vision, audio, language, and other modalities
Keywords:
Remote-Sensing Tiny-Object Detection, Evidence Preservation, Risk-conditioned Interface

Reviews

Among:
Shape:
1234this paper, Bottom 27%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 2.00 over 3 reviews, above 27% of the others.
Reviewer_sYpWOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

PAVER targets tiny-object detection in remote-sensing imagery. Its detector modifies the feature pathway so that weak spatial cues are less likely to disappear during deeper processing. A later confidence stage reads P2/P3 ROI features and uses them to alter candidate scores.

CSFM, SCRG, and HRPE form the detector. The Risk-conditioned Interface is fitted after detector training. Its prototype margin is combined with the detector score, while the detected boxes are left unchanged.

Detection is evaluated on VisDrone, RSOD, and DIOR. VisDrone gives the largest gain: 45.9 mAP50 for PAVER versus 40.2 for YOLOv12-s, with mAP50:95 increasing from 24.0 to 28.0. Confidence quality is studied separately on VisDrone test-dev.

The detector experiments are the more convincing part of the paper. The confidence results are positive relative to raw scores, although the comparison with ordinary score calibration is much less clear.

Critical strengths

The VisDrone detection result is strong. The gain remains visible under stricter localization rather than appearing only at IoU 0.50. AP75 increases from 25.8 to 31.4. APS rises from 16.3 to 21.6.

The architectural study is also fairly extensive. HRPE provides the largest single improvement. The paper then compares the proposed backbone replacement against other blocks, and performs similar controlled comparisons for the fusion and pyramid designs.

These experiments are useful because the full detector modifies several locations at once. The matched replacements make it easier to see whether a gain comes only from adding capacity or from the selected architectural choice.

Detection and confidence calibration are evaluated separately. The confidence study begins from fixed detector outputs, so its results are not mixed with the detector's AP gain.

The fitting protocol is also careful in several respects. Calibration uses the validation set. Test-dev annotations are used for evaluation rather than parameter fitting. Bootstrap resampling is carried out at the image level.

Critical weaknesses

The main concern is the evidence for the Risk-conditioned Interface. PAVER improves the calibration point estimates relative to raw detector confidence, but the more informative comparison is with Platt scaling, which is fitted on the same validation labels and is nested in the proposed formulation when the risk coefficient is zero. Against Platt, the reported differences are only about -0.00015 for ECE, -0.00003 for Brier score, and -0.00014 for NLL, and all three paired 95% intervals include zero. Temperature scaling also achieves lower ECE and NLL than PAVER. PAVER has the best Brier point estimate and FP-AUC, but its FP-AUC of 0.94466 differs from the score-only alternatives at 0.94456 by only 0.000100.00010, without an analogous paired interval. The current results therefore establish an improvement over raw confidence more clearly than an additional benefit from ROI evidence beyond standard score calibration.

A second concern is the definition of candidate correctness. TP/FP labels are generated by greedy matching in raw-score order and then kept fixed across calibration variants, although PAVER can change the candidate ranking. For duplicate detections, a different ordering can change which candidate claims an available ground truth, so a candidate labelled FP before recalibration need not receive the same label if matching is repeated afterward. The calibration metrics consequently evaluate confidence against a correctness event defined under the original ranking, whereas AP is recomputed after the scores change. This protocol is reasonable if that fixed event is the intended target, but it is harder to interpret as correctness of the final ranked output. Repeating TP/FP matching after each recalibrated ranking, and reporting how often assignments change, would clarify this issue.

The confidence study is also narrow in both dataset coverage and uncertainty estimation. The Risk-conditioned Interface is evaluated only on VisDrone, although detection experiments are also reported on RSOD and DIOR. A prototype relation learned from P2/P3 ROI features may depend on object scale, background statistics, or acquisition characteristics, so cross-dataset calibration would be important for a component framed in terms of confidence trustworthiness. In addition, the bootstrap resamples test-dev images while holding the fitted normalizer, prototype banks, and calibration parameters fixed. The intervals therefore describe test-image sampling uncertainty conditional on one fitted interface and omit variation from refitting the confidence model. This omission is relevant because the PAVER-versus-Platt differences are already close to zero. The validation protocol is image-disjoint, but scene-disjointness is not established because suitable scene-group metadata are unavailable.

The evidence-preservation interpretation also needs more direct support. Tables 5–7 show that the proposed architectural components improve AP and that HRPE in particular contributes a sizeable gain. These experiments establish the utility of the design, but they do not directly measure whether weak-target information is preserved through the feature pathway. Figure 1 provides qualitative activation examples, yet no quantitative feature-level diagnostic is reported. Measuring target/background separability across stages, or tracking the retention of small-object responses before and after the relevant transformations, would provide a more direct test of the mechanism. Without such evidence, “evidence preservation” remains a plausible explanation for the AP improvement rather than a verified property of the learned representation.

Efficiency is another concern. At 640×640640\times640, PAVER requires 40.0 GFLOPs compared with 21.4 for YOLOv12-s, while detector-core latency on the reported T4 increases from 2.61 to 4.89 ms per image and peak memory from 0.82 to 1.31 GB. These measurements exclude the Risk-conditioned Interface and the operations required by it, so the end-to-end cost of the complete proposed system is not reported. The comparison with YOLOv12-s at 1024 provides useful context because PAVER at 640 obtains slightly better accuracy at lower nominal GFLOPs, but complete latency remains necessary for evaluating the practical cost of the full pipeline.

The cross-dataset detection evidence is also mixed. PAVER performs strongly on RSOD, reaching 96.7 mAP50 and 68.9 mAP50:95, but on DIOR it matches YOLOv12-s at 74.5 mAP50 while trailing it at mAP50:95, 53.0 versus 55.0. The DIOR evaluation partition also participates in checkpoint selection, so this result is better treated as a controlled internal comparison than as a blind external benchmark. Some matched architectural replacements in Table 6 differ by only 0.2–0.3 AP, yet the paper reports a common seed rather than repeated-run variability for these close comparisons. This is less problematic for the large HRPE gain, but it makes the ordering among closely performing alternatives harder to judge.

Overall, the two main contributions are supported at different levels. The detector has sizeable VisDrone gains, stricter-localization improvements, and reasonably extensive component studies. The Risk-conditioned Interface has favorable point estimates, but its incremental advantage over Platt scaling remains statistically unresolved and is evaluated on only one dataset. Because the confidence interface receives substantial emphasis in the title, abstract, and overall framing, this gap affects the strength of the main contribution rather than only an auxiliary experiment.

Questions for authors

Can a paired 95% interval be reported for FP-AUC when comparing PAVER with Platt scaling?

Can the PAVER-versus-Platt bootstrap be repeated while refitting the calibration stage inside each replicate?

How many TP/FP assignments change after PAVER reorders candidate scores?

What happens to ECE, Brier, and NLL if greedy matching is repeated using each method's final score ordering?

Can the Risk-conditioned Interface be evaluated on RSOD or DIOR?

Is there a quantitative feature-level test for the proposed evidence-preservation mechanism?

For the close replacements in Table 6, does the ranking remain the same across independent training runs?

What is the end-to-end latency after the Risk-conditioned Interface is included?

How sensitive are the calibration results to prototype budget KK?

Do P4 or P5 features give a similar calibration result to P2/P3?

Reviewer_rZuKOpenReview ↗
Rating
3
Confidence
2
Informativeness
2
Presentation
2
Soundness
3
Read review
Paper summary

The authors identify the loss of visual evidence for tiny objects during feature extraction and cross-scale propagation in remote-sensing object detection, and propose the PAVER framework with CSFM, SCRG, and HRPE to enhance tiny-object representations, together with a Risk-conditioned Interface for confidence calibration. Experiments on VisDrone, RSOD, and DIOR demonstrate improved tiny-object detection performance, with additional evaluation of confidence calibration.

Critical strengths
  1. The design of motivation and method is quite well-matched. The author proposed that visual evidence of small targets would gradually be lost during feature extraction, cross-scale fusion and propagation. The following several modules are basically designed to address these issues. Therefore, although the entire method has many components, the logic is relatively smooth.
  2. The overall detection performance of PAVER has improved significantly and the improvement in small target detection is relatively stable.
  3. The ablation study is quite thorough. Not only did it split the three core detector modules to verify their individual effectiveness, but it also made matched replacements with some functionally similar modules. Therefore, it can convincingly demonstrate that the improvement is not solely due to the addition of modules.
Critical weaknesses

The main weakness lies in the fact that the experimental results cannot adequately support the motivation.

Questions for authors
  1. All detection experiments are based on horizontal bounding boxes. Since oriented object detection is particularly important in remote sensing, it remains unclear whether PAVER can provide similar benefits under the more challenging OBB setting.
  2. The validation scope of the Risk-conditioned Interface is rather limited. Currently, the calibration experiments are mainly conducted on VisDrone, so it is not clear whether this risk based on FP/TP prototypes can work stably in other datasets or under distribution shifts.
  3. The computational cost increases substantially (21.5 → 40.0 GFLOPs compared with YOLOv11-s), while the paper does not clearly attribute the additional computational cost to individual components.
  4. Each individual component is reasonably motivated, but none of them appears to introduce a particularly strong new principle. Overall, the method may be viewed as a collection of incremental modifications organized under a unified evidence-preservation narrative.
Reviewer_7bdwOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

The paper proposes PAVER for tiny object detection. Three key designs, CSFM, SCRG, and HRPE are proposed to preserve cross-scale visual evidence of tiny objects in the detector. The paper also proposes a risk-conditioned interface, which uses high-resolution ROI features to calibrate the confidence of detection candidates. The paper was validated on the VisDrone, RSOD, and DIOR remote sensing datasets, showing a significant improvement in AP for small and tiny objects compared to baselines such as YOLOv12-s. The method trades the accuracy gains by higher computational overhead in inference.

Critical strengths

Calibrating the confidence score for tiny object detection is important.

Detailed ablations of each proposed component are well conducted.

Critical weaknesses
  1. The motivation is unclear and the novelty is limited. The three proposed modules (HRPE, CSFM, SCRG) are engineering incremental, with limited conceptual novelty. HRPE adds P2 feature level to enhance the small object detection, which is not new. The motivation behind these architectural designs is not well elaborated.

  2. Writing and presentation are relatively weak. This article suffers from excessive use of terminology and ambiguous naming. For example, the paper uses the word "evidence" to represent something important for detecting the tiny objects. It sounds like a quantifiable and traceable information entity or new pattern. The paper never provides a definition of "evidence". It seems that the evidences are the features more. Since the evidence is an important concept to this work, it is unclear what kind of evidences are preserved. The second is "state-flow". The paper admits that “state flow” denotes an ordered feature-update sequence, not optical flow, a continuous-time model, or an SSM. Then, why not use some easier words? Another example is the "contour". This word is merely ordinary high-resolution features without explicit contour supervision or extraction, making the naming somewhat misleading. "Risk-conditioned" / "risk proxy" sounds like a rigorous risk assessment and uncertainty quantification. However, it is merely a re-scoring based on similarity. This paper relies too heavily on AI writing, and Figure 4 has not been carefully checked by humans.

  3. Too many mathematical symbols are used. Some are repeatedly defined. Thus, the mathematical presentation is not rigorous and uniform. For example, the rr in line 157 denotes risk proxy while it is reference–evolution allocation ratio in line 197. The variable CC denotes the feature maps in Section 3.1 while it is channel width in Section 3.2.

Overall, the paper is mainly about technical contribution on tiny object detection with limited new academical insights. The manuscript relies heavily on AI generation, resulting in low readability.

Questions for authors

The paper needs a major revision. It could have been written in a way that was more fluent and better understood by the community.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.