Pith. sign in

REVIEW 3 major objections 5 minor 61 references

Edge-Aware and Content-Adaptive Infrared Gas Leak Detection for Industrial Safety Monitoring

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Faint, semi-transparent infrared gas plumes can be detected more accurately when the network models gas transport and weak edges, not just generic object features.

desk verdict Useful incremental detector paper with a plausible core result but a reproducibility wall: no code, no seeds, and manual LangGas labels keep the generalization claim from being independently checkable. read the letter →

arxiv 2512.23234 v3 pith:BKLQK74J submitted 2025-12-29 cs.CV cs.AI

Dongsheng Li , Tianli Ma , Siling Wang , Beibei Duan , Song Gao This is my paper
classification cs.CVcs.AI
keywords infraredgasleakdetectionobjectweakplumeedge-awarerepresentationdiffusion-convectionmodelingcontent-adaptiveroutingfeaturepyramidfusionindustrialsafetymonitoring
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to show that infrared gas plumes—faint, small, semi-transparent, and weakly bounded—can be detected more reliably than generic object detectors allow, if the network is built around the physics of gas transport and the visual structure of weak edges. It proposes PEG-DRNet, which combines a diffusion–convection-inspired Gas Block, an adaptive gradient-and-phase edge operator feeding multi-scale edge maps, and a content-adaptive sparse routing neck. The reported experiments on the IIG dataset show 29.8% AP, 84.3% AP50, and 25.3% small-object AP, gains of 3.0, 6.5, and 5.3 points over the RT-DETR-R18 baseline, with less compute (43.7 Gflops, 14.9M params); on the LangGas dataset the method reports 36.3% AP and 68.5% AP50. If these results hold, the practical consequence is an early-warning vision component for industrial gas-leak monitoring that is both more accurate and lighter than current detectors.

What carries the argument

The Gas Block uses a local branch with depthwise convolution as a learned stand-in for the diffusion term D∇²u, and a global branch that applies a learnable frequency-decay kernel in the DCT domain, e^(−αK²), as a stand-in for convection-like long-range transport; edge-aware gating then modulates the global branch. The AGPEO/MSEPM combines directional gradient kernels at 0°, 45°, 90°, and 135° with a phase-congruency response through a learnable scalar α, then downsamples the result into hierarchical edge maps. The CASR-PAN uses an importance estimator to produce per-pixel routing weights such that each cross-scale fusion is a convex combination (1−W)F_local + W F_transport, which the paper

What would settle it

Run PEG-DRNet against RT-DETR-R18 and the strongest YOLO baselines on the same code base, with per-model hyperparameter search and several seeds, reporting mean and variance of AP and AP50; on LangGas, have independent annotators re-label a subset and measure agreement. If the advantage over the strongest baseline collapses to within noise, the paper's central claim fails.

Watch

Extended reading notes

Core claim

The central claim is that the usual bottleneck for infrared gas-leak detection is feature representation, not detector head capacity: faint plumes are missed because generic backbones and static fusion paths cannot capture weak boundaries or long-range plume continuity. PEG-DRNet addresses this with three coupled modules—the Gas Block, which approximates local diffusion with depthwise convolutions and global convection with DCT-domain frequency decay; the adaptive gradient and phase edge operator (AGPEO) feeding a multi-scale edge perception module (MSEPM); and a content-adaptive sparse routing path aggregation network (CASR-PAN) that weights cross-scale feature flow by an importance estimat

Load-bearing premise

The load-bearing premise is that every baseline was given a fair, equally tuned evaluation on the same splits and that the hand-made LangGas boxes are unbiased labels; if either is false, the reported 3–6.5 point improvements are not established.

Editorial extensions

If this is right

  • Weak-plume detection improves without extra compute: 3.0-point AP and 6.5-point AP50 gains over RT-DETR-R18 on IIG come with 13.2 fewer Gflops and about 5M fewer parameters.
  • The larger AP50 gain than AP gain suggests the model's main strength is region-level recall of diffuse, semi-transparent plumes; high-IoU localization (AP75 = 8.5% on IIG) remains the hard part and is acknowledged as future work.
  • The benefit transfers across dataset appearances: on LangGas, PEG-DRNet reaches 36.3% AP and 68.5% AP50, above every compared YOLO, SSD, Faster R-CNN, and RT-DETR baseline.
  • Ablation and routing-path-removal experiments imply each of the three design choices contributes: removing any CASR-PAN path degrades performance, and the full model outperforms PANet, BiFPN, and NAS-FPN necks in the paper's comparisons.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the mechanism is doing what the physics analogy suggests, routing weights should correlate with plume motion in image sequences; that is testable on video data, and the paper's current static-image evaluation does not establish it.
  • The edge-aware modules may transfer to other weak-boundary detection tasks such as smoke, steam, or thermal small targets, but the paper does not test this; a cross-task check would reveal whether the gains come from edge priors generally or from gas-specific training.
  • Because LangGas's detection labels were created for this study, an independent re-annotation or an evaluation using the dataset's original pixel-level masks converted to boxes would be a sharper test of the claimed generalization.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes a single-stage infrared gas-leak detector, called PEG-DRNet in the full text (ECAF-Det in the abstract and highlights), built on RT-DETR. Three modules are introduced: (i) a Gas Block that splits features into a diffusion-like local branch and a DCT-based convection-like global branch with edge-gated fusion; (ii) the AGPEO/MSEPM edge-perception module that fuses directional gradients and a phase-consistency term into multi-scale edge maps; and (iii) CASR-PAN, a content-adaptive sparse routing neck whose importance estimator produces per-location path weights. Experiments on the IIG dataset report 29.8% AP, 84.3% AP50, and 25.3% APS, exceeding RT-DETR-R18 by 3.0, 6.5, and 5.3 points at 43.7 GFLOPs/14.9M parameters; experiments on LangGas report 36.3% AP and 68.5% AP50, exceeding the same baseline by 4.9 points on both. Ablations cover each module, gas-block placement, gradient directions, the AGPEO fusion weight, edge operators, and routing paths.

Significance. If the reported numbers are reproducible, the work is a useful engineering contribution to a niche but safety-relevant detection task. The ablations are internally consistent in direction and unusually thorough for this area: they include stage-wise Gas Block deployment, ERF visualizations, evolution of the learned fusion weight, and a routing-path ablation. The paper is also honest about the limits of its physics analogy: Sec. 3.2 explicitly states that the depthwise convolution is not a Laplacian and that the DCT decay is only an approximation, and Sec. 3.4 labels Eq. (23)-(25) as an analogy. The central weakness is not the method but the evidence for the empirical claim: the LangGas generalization rests entirely on author-created labels with no protocol or release, and no ablation reports variance, seeds, or code. The two-dataset claim, which is the paper's main selling point, therefore cannot currently be independently verified.

major comments (3)
  1. [Sec. 4.1, Sec. 7] The LangGas evaluation is load-bearing for the generalization claim, yet the labels were 'manually labeled' by the authors with no annotation protocol, no inter-annotator agreement, and no release (Sec. 7 only says 'available on request'). If the boxes encode a particular plume-extent convention, every compared method is scored against a label set that may favor the proposed edge/phase architecture. Please provide a detailed labeling protocol, release the labels or at least a reproducible derivation from the existing segmentation annotations, and report inter-annotator agreement. Without this, the 4.9-point LangGas gain cannot be distinguished from label bias.
  2. [Sec. 4.2, Tables 1-9] The evaluation protocol omits the information needed to assess whether the reported gains are meaningful. No random seeds, number of runs, error bars, or statistical significance tests are given; the only training hyperparameters are optimizer, epochs, batch size, and input size. Baseline tuning is not described, so it is unclear whether the compared methods received comparable effort (e.g., the YOLO models may be used with default augmentations while PEG-DRNet uses the RT-DETR recipe). Please report multi-seed mean±std for at least the main comparison and ablations, and release code/weights or provide a detailed per-baseline configuration so the 3.0-6.5 point differences can be independently reproduced.
  3. [Sec. 3.4, Eq. (21)-(23)] There is a mismatch between the formal module definitions and the physical interpretation. AIMM-F is defined in Eq. (21) as Y = F1 + F2·W·(BA+σ(std(F2))), and AIMM-S in Eq. (22) as Y = F·(IDAS+W·(BA+σ(std(F)))), but Eq. (23) describes the output as a convex combination (1-W)Flocal + W·Ftransport. These are different functional forms; the convex-combination reading does not follow from Eqs. (21)-(22). Since the paper uses Eq. (23) to justify the 'spatially varying velocity' analogy, the analogy should be labeled as purely conceptual, and the actual equations should be used for the routing interpretation. This does not invalidate the empirical results, but it is a correctness issue in the method's presentation.
minor comments (5)
  1. [Title/Abstract vs. Full Text] The abstract and highlights name the model ECAF-Det while the full text uses PEG-DRNet; Figure 6's caption also calls it GASRNet. Please unify the name throughout.
  2. [Sec. 4.5.6, Table 6] The ablation of gradient directions shows the 4-direction variant has lower AP (27.9) and AP50 (80.4) than the 2-direction variant (28.8 AP, 81.1 AP50), yet the text claims multi-directional modeling is 'critical for high-precision edge detection.' The evidence supports the AP75/APS claim but not the blanket conclusion; please qualify the claim or provide an explanation grounded in the precision-recall trade-off.
  3. [Sec. 4.5.2, Fig. 8] The ERF comparison reports 'high-contribution area ratio' but the text's sentence 'the high-contribution area ratio of Gas Block reaches 0.047 and 0.340' reads as if 0.340 is larger than 0.047 for the same threshold; please clarify which threshold corresponds to which value and add axis labels to Figure 8, which currently contains placeholder glyphs.
  4. [Sec. 4.3, Table 1] The text says 'Notably, PEG-DRNet attains a substantial improvement in small-object detection (APS=25.3%, +5.3% over Yolov12n)' but the table shows +5.4 over Yolov12n and +5.3 over RT-DETR-R18. Please make the baseline explicit in the sentence.
  5. [Sec. 4.5.9, Table 8] The 'Naive additive fusion' row has AP 25.9, which is lower than the 'Without deep-to-mid fusion' row (27.5) and 'Without mid-level self-fusion' (28.3), but the text says 'removing any individual routing path leads to a consistent performance degradation compared with the full CASR-PAN' — that is true, but the naive baseline being much lower than some ablated variants is not discussed. Please add a sentence explaining why removing a path can improve over simple addition.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: reported gains are empirical train/test measurements, and the physics equations are explicitly motivational analogies, not fitted predictors.

full rationale

The paper's central claim is an evaluated detector (PEG-DRNet) with measured AP/AP50 on IIG and LangGas. These numbers come from training and testing, not from a derivation chain that returns its inputs. The diffusion-convection equation (Eq. 1) and its Fourier solution (Eq. 5) motivate the Gas Block, but the paper explicitly disclaims exactness: 'a learnable depthwise convolution does not strictly correspond to a discrete Laplacian' and 'the proposed global branch does not explicitly implement this phase behavior.' Similarly, CASR-PAN's routing equation (23) is introduced as an analogy ('The structure of Eq. (23) mirrors the explicit discretization of the convection operator') and Eq. (25) is explicitly approximate ('By analogy'), so no output is forced by definition. The ablations compare actual ablated architectures under otherwise identical settings; even if some training details or the manually created LangGas labels are undocumented, that is a reproducibility/validity concern, not circularity. No load-bearing premise is justified solely by a self-citation chain—the dataset citations (Yu et al. 2024; Guo et al. 2025) are external sources. The abstract/full-text name discrepancy (ECAF-Det vs PEG-DRNet) is a consistency issue, not a circular derivation.

Assumptions & free parameters 5 free parameters · 4 assumptions · 1 invented entities

The paper's central claim rests on a small number of hand-set or learned scalar modulation constants, two benchmark datasets, and a loose physical analogy that the authors themselves describe as approximate. No new physical entity is introduced; one interpretive 'velocity field' analogy is flagged as non-independent.

free parameters (5)
  • α (Gas Block frequency decay coefficient) = learned, exact value not reported
    Eq. 9 uses e^{-αK²} to emulate diffusion; α is a learnable surrogate for Dt, trained end-to-end.
  • α (AGPEO gradient-phase fusion weight) = initialized 0.7, converges to ≈0.77–0.79 (Fig. 13)
    Eq. 16 fuses gradient and phase maps with a learnable scalar; ablation shows sensitivity to this choice.
  • BA (bias addition lower bound) = 0.5 (fixed)
    Eq. 21/22 use BA=0.5 as a hand-set constant to prevent over-suppression during adaptive fusion.
  • IDAS (identity-aware scaling) = 1 (fixed)
    Eq. 22 sets IDAS=1 by design to preserve an identity mapping in self-enhancement.
  • w_g, w_l, w_d (importance estimator fusion weights) = learned, values not reported
    Eq. 19 combines global/local/diversity importance with three learnable weights normalized by softmax.
assumptions (4)
  • domain assumption Gas plume appearance in IR video is governed by the convection-diffusion equation (Eq. 1) and can be usefully emulated by local DWConv + DCT decay.
    Section 3.2 invokes Eq. 1 to justify the Gas Block; the paper concedes the emulation is approximate.
  • domain assumption The IIG and LangGas bounding-box annotations correctly localize ambiguous, semi-transparent gas plumes; manual LangGas labels are unbiased.
    Section 4.1: LangGas labels were manually created by the authors with no protocol or agreement metric.
  • domain assumption COCO-style AP, AP50, AP75, and scale-split metrics on these two datasets are a fair and sufficient measure of gas leak detection quality.
    Section 4.2 adopts 'COCO evaluation protocol'; AP75 is very low (~8–12) and the paper argues AP50 is more meaningful—an assumption about which metric matters.
  • ad hoc to paper Backbone features can be split into diffusion-like (local) and convection-like (global) components that do not need explicit phase modeling.
    Section 3.2: phase-shift convection term is omitted because DCT is real-valued; edge gating is said to act as an implicit surrogate.
invented entities (1)
  • Implicit spatially varying 'velocity magnitude' W(x,y)≈∥v(x,y)∥Δt
    purpose: Interpretability metaphor for content-adaptive routing weights in CASR-PAN (Eq. 23–25); not a measured physical field.
    The paper equates routing weights to a convection velocity solely by analogy; no velocity measurements or physics validation are provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Edge-Aware and Content-Adaptive Infrared Gas Leak Detection for Industrial Safety Monitoring." pith.science (2026). https://pith.science/paper/BKLQK74J

@misc{pith2026251223234,
  author       = {Pith},
  title        = {Pith review of: Edge-Aware and Content-Adaptive Infrared Gas Leak Detection for Industrial Safety Monitoring},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BKLQK74J}},
  note         = {Machine review of arXiv:2512.23234}
}
read the original abstract

Infrared gas leak detection is important for industrial safety and environmental monitoring, but automatic detection remains challenging because gas plumes are often faint, small, semi-transparent, and weakly bounded. This paper proposes an Edge-Aware and Content-Adaptive Feature Fusion Detector (ECAF-Det) for weak-plume detection in cluttered thermal scenes. ECAF-Det integrates three task-oriented designs: a plume-oriented local-global feature enhancement block to preserve fine boundary cues and capture long-range contextual continuity; a multi-scale edge perception module that transforms directional gradient and phase-consistency cues into hierarchical edge priors for boundary-sensitive plume representation; and a content-adaptive sparse routing path aggregation network that dynamically regulates multi-scale feature propagation to emphasize informative plume features and suppress redundant background responses. Experiments on the IIG dataset show that ECAF-Det achieves 29.8% AP, 84.3% AP50, and 25.3% small-object AP, improving the RT-DETR-R18 baseline by 3.0, 6.5, and 5.4 percentage points, respectively, with 43.7 GFLOPs and 14.9 M parameters. On the LangGas dataset, ECAF-Det achieves 36.3% AP and 68.5% AP50, demonstrating its generalization to different infrared gas plume appearances. The main AI contribution is edge-aware representation learning with content-adaptive sparse feature routing for weak infrared plume perception. The proposed detector can serve as a visual perception component for early warning and remote inspection in industrial gas leak monitoring.

Figures

Figures reproduced from arXiv: 2512.23234 by the authors.

Figure 1
Figure 1. Comparing the AP50 and Gflops of detection methods on IIG. discriminability while significantly reducing redundant computation. • Comprehensive validation and interpretability. We conducted extensive experiments on the IIG and LangGas datasets. PEG-DRNet achieves superior AP50 performance compared with both CNN- and Transformer￾based detectors. It also provides physically interpretable feature behav￾iors that align … view at source ↗
Figure 2
Figure 2. The overall architecture of our proposed PEG-DRNet. In AIMM-F, the bias [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. The Physics–Edge Hybrid Backbone integrates initial convolution and down [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Architecture of the multi-scale edge perception module (MSEPM) and adaptive [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Architecture of the importance estimator. [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 6
Figure 6. Figure 6: Visualization comparison of different detectors on the IIG dataset. Each row [PITH_FULL_IMAGE:figures/full_fig_p024_6.png]
Figure 7
Figure 7. Figure 7: Effective receptive field (ERF) visualization comparison between (a) the pro [PITH_FULL_IMAGE:figures/full_fig_p028_7.png]
Figure 8
Figure 8. Figure 8: Comparison of high-contribution area ratios between the proposed Physics-Edge [PITH_FULL_IMAGE:figures/full_fig_p029_8.png]
Figure 9
Figure 9. Figure 9: Performance comparison of different backbones. The red cross indicates the [PITH_FULL_IMAGE:figures/full_fig_p030_9.png]
Figure 10
Figure 10. Figure 10: Visualization results of the proposed AGPEO module. (a) Original infrared [PITH_FULL_IMAGE:figures/full_fig_p031_10.png]
Figure 11
Figure 11. Figure 11: Visualization of multi-scale edge extraction results generated by the proposed [PITH_FULL_IMAGE:figures/full_fig_p032_11.png]
Figure 12
Figure 12. Figure 12: Effect of the fusion weight α in AGPEO. Performance comparisons under α = 0, α = 1, and learnable α are reported in terms of (a) AP, (b) AP50, and (c) APS. We conduct an ablation study on the fusion weight α within the AG￾PEO module of PEG-DRNet [PITH_FULL_IMAGE:figu…
Figure 13
Figure 13. Figure 13: Training evolution of the learnable fusion weight [PITH_FULL_IMAGE:figures/full_fig_p035_13.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

61 extracted references · 10 canonical work pages

  1. [1]

    Bonvicini, S., Antonioni, G., Morra, P., & Cozzani, V. (2015). Quantita- tive assessment of environmental risk due to accidental spills from on- shore pipelines.Process Safety and Environmental Protection, 93, 31– 49.https://doi.org/10.1016/j.psep.2014.04.007. 38

    work page doi:10.1016/j.psep.2014.04.007 2015

  2. [2]

    Canny, J. (1986). A computational approach to edge detection.IEEE Transactions on Pattern Analysis and Machine Intelligence, 8, 679– 698.https://doi.org/10.1109/TPAMI.1986.4767851

    arXiv 1986

  3. [3]

    Zagoruyko, S. (2020). End-to-end object detection with transform- ers. InEuropean Conference on Computer Vision(pp. 213–229). Springer.https://doi.org/10.1007/978-3-030-58452-8_13

    doi:10.1007/978-3-030-58452-8_13 2020

  4. [4]

    Chen, H., Wang, Y., Guo, J., & Tao, D. (2023). VanillaNet: The power of minimalism in deep learning.Advances in Neural Information Processing Systems, 36, (pp. 7050–7064)

    2023

  5. [5]

    Chen, X., Li, X., Fu, B., Hou, X., Gan, W., & Huang, C. (2024). FBG strain sensing technology-based gas pipeline leak mon- itoring and accurate location.Engineering Failure Analysis, 159, 108102.https://doi.org/10.1016/j.engfailanal.2024.108102

    arXiv 2024

  6. [9]

    Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014). Rich feature hier- archies for accurate object detection and semantic segmentation. In2014 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 580–587).https://doi.org/10.1109/CVPR.2014.81. 39

    doi:10.1109/cvpr.2014.81 2014

  7. [10]

    Girshick, R. (2015). Fast R-CNN. In2015 IEEE Interna- tional Conference on Computer Vision (ICCV)(pp. 1440– 1448).https://doi.org/10.1109/ICCV.2015.169

    doi:10.1109/iccv.2015.169 2015

  8. [11]

    Ghiasi, G., Lin, T.-Y., & Le, Q. V. (2019). NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection. In2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(pp. 7029–7038).https://doi.org/10.1109/CVPR.2019.00720. [dataset]Guo, W. M., & Du, Y. (2025). LangGas: Introducing language in selective zero-shot background ...

    arXiv 2019

Show all 61 references
  1. [12]

    Heath, M., Sarkar, S., Sanocki, T., & Bowyer, K. (1998). Comparison of edge detectors: A methodology and initial study.Computer Vision and Image Understanding, 69, 38–54.https://doi.org/10.1006/cviu.1997.0587

    1998

  2. [13]

    Han, Y., Feng, X., & Todd, M. D. (2023). A novel method- ology for quantitative identification of pipeline leakage and nega- tive pressure wave velocity.Structural Health Monitoring, 22, 2267– 2279.https://doi.org/10.1177/14759217221123403

    2023 doi

  3. [14]

    Jiang, P., Ergu, D., Liu, F., Cai, Y., & Ma, B. (2022). A review of YOLO algorithm developments.Procedia Computer Science, 199, 1066– 1073.https://doi.org/10.1016/j.procs.2022.01.135

    2022 doi

  4. [15]

    Jing, Y., Sun, Y., & Wang, Q. (2025). Lightweight single- stage network for gas leak detection based on infrared imag- ing.IEEE Transactions on Instrumentation and Measurement, 7, 1–9.https://doi.org/10.1109/TIM.2025.3561424

    2025

  5. [16]

    Kopbayev, A., Khan, F., Yang, M., & Halim, S. Z. (2022). Gas leakage detection using spatial and temporal neural network model.Process Safety and Environmental Protection, 160, 968– 975.https://doi.org/10.1016/j.psep.2022.03.002

    2022 doi

  6. [17]

    Kang, Z., Qian, X., Li, Y., Hou, L., Huang, Z., Duanmu, W., & Yuan, M. (2023). Feature extraction of natural gas leak- age for an intelligent warning model: A data-driven analysis and 40 modeling.Process Safety and Environmental Protection, 174, 574– 584.https://doi.org/10.101...

    2023 doi

  7. [18]

    Khanam, R., & Hussain, M. (2024). YOLOv11: An overview of the key architectural enhancements.arXiv preprint arXiv:2410.17725.https://arxiv.org/abs/2410.17725

    2024 arXiv

  8. [19]

    Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014). Microsoft COCO: Common objects in context. InEuropean Conference on Computer Vision(pp. 740–755). Springer.https://doi.org/10.1007/978-3-319-10602-1_48

    2014 doi

  9. [20]

    Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.-Y., & Berg, A. C. (2016). SSD: Single shot multibox detector. InEuropean Confer- ence on Computer Vision(pp. 21–37).https://doi.org/10.1007/978-3-319- 46448-0_2

    2016 doi

  10. [21]

    Lin, T.-Y., Dollár, P., Girshick, R., He, K., Hariharan, B., & Belongie, S. (2017). Feature pyramid networks for object detection. In2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)(pp. 936–944).https://doi.org/10.1109/CVPR.2017.106

    2017 doi

  11. [22]

    (2018, June)

    Liu, S., Qi, L., Qin, H., Shi, J., & Jia, J. (2018, June). Path aggre- gation network for instance segmentation. In2018 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR)(pp. 8759–8768).https://doi.org/10.1109/CVPR.2018.00913

    2018

  12. [23]

    Lu, H., Iseley, T., Behbahani, S., & Fu, L. (2020). Leak- age detection techniques for oil and gas pipelines: State-of- the-art.Tunnelling and Underground Space Technology, 98, 103249.https://doi.org/10.1016/j.tust.2019.103249

    2020

  13. [24]

    M., & Zhang, L

    Li, F., Zhang, H., Liu, S., Guo, J., Ni, L. M., & Zhang, L. (2022). DN-DETR: Accelerate DETR Training by Introduc- ing Query DeNoising. In2022 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR)(pp. 13609– 13617).https://doi.org/10.1109/CVPR52688.2022.01325

    2022

  14. [26]

    Li, Y., Hu, J., Wen, Y., Evangelidis, G., Salahi, K., Wang, Y., Tulyakov, S., & Ren, J. (2023). Rethinking vision trans- formers for MobileNet size and speed. In2023 IEEE/CVF International Conference on Computer Vision (ICCV)(pp. 16843–16854).https://doi.org/10.1109/ICCV51070....

    2023

  15. [27]

    Li, H., Zhang, R., Pan, Y., Ren, J., & Shen, F. (2024). LR-FPN: Enhanc- ing remote sensing object detection with location refined feature pyra- mid network. In2024 International Joint Conference on Neural Networks (IJCNN)(pp. 1–8).https://doi.org/10.1109/IJCNN60899.2024.10650583

    2024

  16. [28]

    Li, K., Chen, W., Zou, Y., Wang, Z., Zhou, X., & Shi, J. (2025). Optimized PSOMV-VMD combined with Con- vFormer model: A novel gas pipeline leakage detection method based on low sensitivity acoustic signals.Measurement, 247, 116804.https://doi.org/10.1016/j.measurement.2025.116804

    2025

  17. [29]

    Li, J., Lu, Y., Guo, X., Xian, X., Wang, T., & Shi, Y. (2026). Ex- ploiting Gaussian agnostic representation learning with diffusion pri- ors for enhanced infrared small target detection.Neural Networks, 194, 108185.https://doi.org/10.1016/j.neunet.2025.108185

    2026

  18. [30]

    Wang, J. (2021). Conditional DETR for Fast Training Convergence. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV) (pp. 3631–3640).https://doi.org/10.1109/ICCV48922.2021.00363

    2021

  19. [31]

    Meribout, M. (2021). Gas leak-detection and measurement systems: Prospects and future trends.IEEE Transactions on Instrumentation and Measurement, 70, 1–13.https://doi.org/10.1109/TIM.2021.3096596

    2021

  20. [32]

    Pan, J., Peng, D., Wang, D., Zhao, H., & Qu, B. (2025). SRHS-Net: A hybrid attentive architecture for real-time steam leak detection in power plants.IEEE Transactions on Industrial Informatics, 21(9), 6879– 6890.https://doi.org/10.1109/TII.2025.3568498

    2025

  21. [33]

    B., & Kim, J.-M

    Quy, T. B., & Kim, J.-M. (2022). Pipeline leak detection us- ing acoustic emission and state estimate in feature space. 42 IEEE Transactions on Instrumentation and Measurement, 71, 1–9.https://doi.org/10.1109/TIM.2022.3206833

    2022

  22. [34]

    Qiao, C., Shen, F., Wang, X., Wang, R., Cao, F., Zhao, S., & Li, C. (2022). A novel multi-frequency coordinated module for SAR ship detection. In2022 IEEE 34th International Conference on Tools with Artificial Intelligence (ICTAI)(pp. 804–811).https://doi.org/10.1109/ICTAI5601...

    2022

  23. [35]

    Ren, S., He, K., Girshick, R., & Sun, J. (2017). Faster R-CNN: To- wards real-time object detection with region proposal networks.IEEE Transactions on Pattern Analysis and Machine Intelligence, 39, 1137– 1149.https://doi.org/10.1109/TPAMI.2016.2577031

    2017

  24. [36]

    Strahl, T., Herbst, J., Lambrecht, A., Maier, E., Steinebrunner, J., & Wöllenstein, J. (2021). Methane leak detection by tunable laser spectroscopy and mid-infrared imaging.Applied Optics, 60, C68– C75.https://doi.org/10.1364/AO.419942

    2021 doi

  25. [37]

    Shen, F., & Tang, J. (2024). ImagPose: A unified conditional framework for pose-guided person generation.Advances in Neural Information Processing Systems, 37, 6246–6266

    2024

  26. [38]

    Shen, F., Ye, H., Zhang, J., Wang, C., Han, X., & Wei, Y. (2024). Advancing pose-guided image synthesis with progressive conditional diffusion models. InProceedings of the Twelfth International Conference on Learning Rep- resentations

    2024

  27. [39]

    Shen, F., Xu, W., Yan, R., Zhang, D., Shu, X., & Tang, J. (2025). IMAGEdit: Let any subject transform.arXiv preprint arXiv:2510.01186.https://arxiv.org/abs/2510.01186

    2025 arXiv

  28. [40]

    Tang, J. (2025). Imagdressing-v1: Customizable virtual dressing. InPro- ceedings of the AAAI Conference on Artificial Intelligence, 39, 6795– 6804.https://doi.org/10.1609/aaai.v39i7.32729

    2025 doi

  29. [41]

    Shen, F., Ye, H., Liu, S., Zhang, J., Wang, C., Han, X., & Wei, Y. (2025). Boostingconsistencyinstoryvisualizationwithrich-contextualconditional diffusion models. InProceedings of the AAAI Conference on Artificial In- telligence, 39, 6785–6794.https://doi.org/10.1609/aaai.v39i...

    2025 doi

  30. [42]

    Tang, J. (2025). IMAGHarmony: Controllable image edit- ing with consistent object quantity and layout.arXiv preprint arXiv:2506.01949.https://arxiv.org/abs/2506.01949

    2025

  31. [43]

    Shen, F., Yu, J., Wang, C., Jiang, X., Du, X., & Tang, J. (2025). IMAGGarment-1: Fine-grained garment gen- eration for controllable fashion design.arXiv preprint arXiv:2504.13176.https://arxiv.org/abs/2504.13176

    2025 arXiv

  32. [44]

    Tan, M., & Le, Q. (2019). EfficientNet: Rethinking model scaling for con- volutional neural networks. In K. Chaudhuri & R. Salakhutdinov (Eds.), Proceedings of the 36th International Conference on Machine Learning (pp. 6105–6114)

    2019

  33. [45]

    Tan, M., Pang, R., & Le, Q. V. (2020). EfficientDet: Scalable and effi- cient object detection. InProceedings of the 2020 IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR)(pp. 10778– 10787).https://doi.org/10.1109/CVPR42600.2020.01079

    2020

  34. [46]

    Tian, Y., Ye, Q., & Doermann, D. (2025). YOLOv12: Attention-centric real-time object detectors.arXiv preprint arXiv:2502.12524.https://doi.org/10.48550/arXiv.2502.12524

    2025 doi arXiv

  35. [47]

    A., Homer, M

    Vergara, A., Vembu, S., Ayhan, T., Ryan, M. A., Homer, M. L., & Huerta, R. (2012). Chemical gas sensor drift compensation using classifier ensembles.Sensors and Actuators B: Chemical, 166, 320– 329.https://doi.org/10.1016/j.snb.2012.01.074

    2012 doi

  36. [48]

    Wang, X. (2007). Laplacian operator-based edge detectors.IEEE Transactions on Pattern Analysis and Machine Intelligence, 29, 886–890.https://doi.org/10.1109/TPAMI.2007.1027

    2007

  37. [49]

    Zimmerle, D., Savarese, S., & Brandt, A. R. (2020). Machine vision for natural gas methane emissions detection using an infrared camera.Applied Energy, 257, 113998.https://doi.org/10.1016/j.apenergy.2019.113998

    2020

  38. [50]

    P., Savarese, S., & Brandt, A

    Wang, J., Ji, J., Ravikumar, A. P., Savarese, S., & Brandt, A. R. (2022). VideoGasNet: Deep learning for natural gas 44 methane leak classification using an infrared camera.Energy, 238, 121516.https://doi.org/10.1016/j.energy.2021.121516

    2022

  39. [51]

    K., Goel, S., Lin, Z., & Naidu, R

    Wang, L., Cheng, Y., Gopalan, S., Luo, F., Amreen, K., Singh, R. K., Goel, S., Lin, Z., & Naidu, R. (2023). Review and per- spective: Gas separation and discrimination technologies for current gas sensors in environmental applications.ACS Sensors, 8, 1373– 1390.https://doi.org...

    2023 doi

  40. [52]

    S., & Xie, S

    Woo, S., Debnath, S., Hu, R., Chen, X., Liu, Z., Kweon, I. S., & Xie, S. (2023). ConvNeXt V2: Co-designing and scaling Con- vNets with masked autoencoders. In2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(pp. 16133– 16142).https://doi.org/10.1109/CV...

    2023

  41. [53]

    Wang, A., Chen, H., Liu, L., Chen, K., Lin, Z., Han, J., & others. (2024). Yolov10: Real-time end-to-end object detection.Advances in Neural In- formation Processing Systems, 37, 107984–108011

    2024

  42. [55]

    Weng, W., Wei, M., Ren, J., & Shen, F. (2024). Enhanc- ing aerial object detection with selective frequency interaction network.IEEE Transactions on Artificial Intelligence, 5(12), 6109–6120.https://doi.org/10.1109/TAI.2024.3381096

    2024

  43. [56]

    Wang, F., Zhao, H., Peng, Y., Fang, J., Liu, P., & Zhang, R. (2026). MI- IGAN: Mambas make strong GAN for infrared image generation.Neural Networks, 193, 108021.https://doi.org/10.1016/j.neunet.2025.108021

    2026

  44. [57]

    Xu, G., Chen, Y., Yang, M., Li, S., & Marma, K. J. S. (2023). An outlook analysis on China’s natural gas consumption forecast by 2035: Applying a seasonal forecasting method.Energy, 284, 128602.https://doi.org/10.1016/j.energy.2023.128602. [dataset]Yu, H., Wang, J., Wang, Z., ...

    2023

  45. [58]

    Yao, J., Xiong, Z., Li, S., Yu, Z., & Liu, Y. (2024). TSFF- Net: A novel lightweight network for video real-time detec- tion of SF6 gas leaks.Expert Systems with Applications, 247, 123219.https://doi.org/10.1016/j.eswa.2024.123219

    2024

  46. [59]

    Yao, L., Zhang, L., Wang, L., Li, R., & Luo, H. (2025). A spatiotemporal data-driven framework for acoustic signal-based natural gas pipeline leak detection.IEEE Transactions on Instrumentation and Measurement, 74, 1–12.https://doi.org/10.1109/TIM.2025.3556201

    2025

  47. [60]

    Yu, W., & Wang, X. (2025). MambaOut: Do we really need Mamba for vision? In2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(pp. 4484–4496).https://doi.org/10.1109/CVPR52734.2025.00423

    2025

  48. [61]

    Zimmerle, D., Vaughn, T., Bell, C., Bennett, K., Deshmukh, P., & Thoma, E. (2020). Detection limits of optical gas imaging for natural gas leak de- tection in realistic controlled conditions.Environmental Science & Tech- nology, 54, 11506–11514.https://doi.org/10.1021/acs.est.0c01285

    2020 doi

  49. [62]

    Zhao, L., Cao, Z., & Deng, J. (2024). A review of leak detection methods based on pressure waves in gas pipelines.Measurement, 236, 115062.https://doi.org/10.1016/j.measurement.2024.115062

    2024

  50. [63]

    Zhong, S., Zhou, H., Zheng, Z., Ma, Z., Zhang, F., & Duan, J. (2024). Hierarchical attention-guided multiscale aggregation net- work for infrared small target detection.Neural Networks, 171, 485– 496.https://doi.org/10.1016/j.neunet.2023.12.036

    2024 doi

  51. [64]

    Liu, Y., & Chen, J. (2024). DETRs beat YOLOs on real- time object detection. In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(pp. 16965–16974).https://doi.org/10.1109/CVPR52733.2024.01605. 46

    2024

  52. [65]

    Zuo, Z., Zhang, H., Ma, L., Liu, T., & Liang, S. (2024). Leak detection for natural gas gathering pipelines under mul- tiple operating conditions using RP-1dConvLSTM-AE and multi- model decision.IEEE Transactions on Industrial Electronics, 71, 6263–6273.https://doi.org/10.1109...

    2024

  53. [66]

    Zhang, F., Hu, H., Zou, B., & Luo, M. (2025). M4Net: Multi- level multi-patch multi-receptive multi-dimensional attention net- work for infrared small target detection.Neural Networks, 183, 107026.https://doi.org/10.1016/j.neunet.2024.107026. 47

    2025

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.