pith. machine review for the scientific record. sign in

arxiv: 2605.05627 · v1 · submitted 2026-05-07 · 💻 cs.CV · cs.AI· cs.LG· cs.RO

Recognition: unknown

Leveraging Image Generators to Address Training Data Scarcity: The Gen4Regen Dataset for Forest Regeneration Mapping

Gabriel Jeanson , David-Alexandre Duclos , William Larriv\'ee-Hardy , No\'e Cochet , Mat\v{e}j Boxan , Anthony Desch\^enes , Fran\c{c}ois Pomerleau , Philippe Gigu\`ere
Authors on Pith no claims yet

Pith reviewed 2026-05-08 15:04 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.LGcs.RO
keywords semantic segmentationsynthetic dataUAV imageryforest regenerationspecies mappingdata scarcityvision-language models
0
0 comments X

The pith

Vision-language models generate synthetic images that raise F1 scores for forest species mapping by over 15 percentage points when mixed with limited real UAV data.

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

The paper shows how to tackle the shortage of expert-labeled aerial photos needed to train models that map individual tree species in young forest regrowth zones. A vision-language model creates realistic synthetic images together with matching pixel-level masks from simple text prompts, which are then added to a small collection of real UAV images during training. This combined approach lifts overall segmentation performance and delivers especially large gains on rare species that appear infrequently in the real data. The work matters for sustainable forestry because manual ground surveys cannot scale to large areas and expert annotation of drone photos remains costly and slow. If the method holds, it opens a path to more frequent and detailed species inventories without proportional increases in labeling effort.

Core claim

Training semantic segmentation models on a mixture of real UAV forest images and prompt-generated synthetic images with automatically produced masks produces substantially higher accuracy on real test images than training on real data alone, with overall F1 score gains exceeding 15 percentage points and per-species gains reaching 30 percentage points for underrepresented regeneration species.

What carries the argument

Prompt-driven generation of paired high-resolution images and pixel-aligned semantic masks by a vision-language model, which are mixed with real labeled and unlabeled UAV images to train semantic segmentation networks for fine-grained forest species.

If this is right

  • Models achieve higher accuracy in distinguishing fine-grained regeneration species from aerial views.
  • Small volumes of generated data correct severe class imbalance and raise performance on rare species.
  • Reliance on time-consuming manual photo-interpretation for new training sets is reduced.
  • Vision-language models become practical tools for creating labeled data in other label-scarce remote-sensing domains.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The technique could be tested on other ecological monitoring tasks such as wetland mapping or invasive species detection where real labels are equally scarce.
  • Iterative loops in which model outputs refine the text prompts for further generation rounds might further reduce the need for any real labels.
  • Scaling the approach to larger synthetic pre-training sets followed by minimal real fine-tuning could be evaluated for cost savings in operational forestry programs.

Load-bearing premise

The synthetic images and masks must be realistic enough and free of systematic artifacts so that models trained on them still perform well when tested on actual UAV photographs.

What would settle it

Training a segmentation model on the combined real-plus-synthetic dataset and finding that it achieves the same or lower F1 score than a model trained only on real images when both are evaluated on a held-out set of independent real UAV forest images.

Figures

Figures reproduced from arXiv: 2605.05627 by Anthony Desch\^enes, David-Alexandre Duclos, Fran\c{c}ois Pomerleau, Gabriel Jeanson, Mat\v{e}j Boxan, No\'e Cochet, Philippe Gigu\`ere, William Larriv\'ee-Hardy.

Figure 1
Figure 1. Figure 1: Overview of our model training pipeline, combining 199 manually-labelled, 25 106 pseudo-labelled real images, and 2101 view at source ↗
Figure 2
Figure 2. Figure 2: UAV image (top left) with its corresponding manually labelled semantic segmentation mask (bottom left), compared to a view at source ↗
Figure 3
Figure 3. Figure 3: Class distribution in the labelled training datasets. The labelled dataset of real images exhibits a strong class imbalance, despite view at source ↗
Figure 4
Figure 4. Figure 4: Map of the different data collection locations present in view at source ↗
Figure 5
Figure 5. Figure 5: Proportion of pixels per class seen by the model during training. These statistics are a result of the number of real vs. generated view at source ↗
Figure 6
Figure 6. Figure 6: Per-species F1 score comparing the DINOv2 model trained using only hand-labelled data, versus the approach when training on view at source ↗
Figure 7
Figure 7. Figure 7: Selected qualitative examples using the DINOv2 model, for combinations of training data. view at source ↗
Figure 8
Figure 8. Figure 8: F1 score based on the number of AI-generated images used during training. The DINOv2 model was used, and the 40:60 view at source ↗
read the original abstract

Sustainable forest management relies on precise species composition mapping, yet traditional ground surveys are labour-intensive and geographically constrained. While Uncrewed Aerial Vehicles (UAVs) offer scalable data collection, the transition to deep learning-based interpretation is bottlenecked by the severe scarcity of expert-annotated imagery, particularly in complex, visually heterogeneous regeneration zones. This paper addresses the dual challenges of data scarcity and extreme class imbalance in the semantic segmentation of fine-grained forest regeneration species by providing a scalable framework that reduces reliance on manual photo-interpretation for high-resolution, millimetre-level aerial imagery. Importantly, we leverage the large-scale vision-language Nano Banana Pro model to simultaneously generate high-fidelity images and their corresponding pixel-aligned semantic masks from prompts. We introduce WilDReF-Q-V2, an expansion of a natural forest dataset with 13 977 new unlabelled and 50 labelled real images, as well as the Gen4Regen dataset, featuring 2101 pairs of synthetic images and semantic masks. Our methodology integrates real-world data with AI-generated images, highlighting that AI-generated data is highly complementary to real-world data, with unified training yielding an F1 score improvement of over 15 %pt compared to purely supervised baselines. Furthermore, we demonstrate that even small quantities of prompt-generated data significantly improve performance for underrepresented species, some of which saw per-species F1 score gains of up to 30 %pt. We conclude that vision-language models can serve as agile data generators, effectively bootstrapping perception tasks for niche AI domains where expert labels are scarce or unavailable. Our datasets, source code, and models will be available at https://norlab-ulaval.github.io/gen4regen.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes using the Nano Banana Pro vision-language model to generate paired synthetic images and pixel-aligned semantic masks for fine-grained forest regeneration species segmentation from UAV imagery. It releases the Gen4Regen dataset (2101 synthetic pairs) and expands the real WilDReF-Q-V2 dataset (13,977 unlabeled + 50 labeled images). The central empirical claim is that unified training on real + synthetic data yields >15 percentage point F1 improvement over purely supervised real-data baselines, with per-species gains up to 30 percentage points for underrepresented classes, demonstrating that VLM-generated data is complementary for addressing scarcity and imbalance.

Significance. If the synthetic masks prove accurate and free of domain-specific artifacts, the work offers a practical, scalable route to augment scarce expert labels in niche remote-sensing tasks. Public release of datasets, code, and models is a clear strength supporting reproducibility. The result would be most impactful if accompanied by evidence that gains arise from genuine pattern learning rather than exploitation of synthetic cues.

major comments (2)
  1. [Abstract / Results] Abstract and results sections: the headline claim of >15 %pt F1 gain (and up to 30 %pt on rare species) from unified training is presented without any description of the segmentation architecture, training protocol, test-set composition, number of random seeds, or statistical significance testing. This absence prevents verification that the reported improvements are robust rather than artifacts of a particular split or baseline choice.
  2. [Methodology / Dataset Generation] Dataset generation and methodology: no quantitative validation is provided that the automatically generated semantic masks are pixel-accurate on real UAV distributions (e.g., no held-out mask audit, boundary precision metrics, or label-distribution distance between synthetic and expert annotations). Without such checks, the observed gains could stem from the model learning VLM-specific textures or hallucinated boundaries rather than true regeneration patterns, directly undermining the central claim that the synthetic data is “highly complementary.”
minor comments (2)
  1. Clarify whether “Nano Banana Pro” is a publicly available model with fixed weights or a custom/fine-tuned variant; if the latter, release the exact prompt templates and generation parameters to enable reproduction.
  2. [Abstract] The abstract states that “even small quantities of prompt-generated data significantly improve performance”; provide the exact quantities used in the ablation and the corresponding per-species F1 tables to support this statement.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and valuable comments on our manuscript. We provide point-by-point responses to the major comments below and outline the revisions we will make to address the concerns raised.

read point-by-point responses

  1. Referee: [Abstract / Results] Abstract and results sections: the headline claim of >15 %pt F1 gain (and up to 30 %pt on rare species) from unified training is presented without any description of the segmentation architecture, training protocol, test-set composition, number of random seeds, or statistical significance testing. This absence prevents verification that the reported improvements are robust rather than artifacts of a particular split or baseline choice.

    Authors: We agree that the abstract and results sections would benefit from additional details to support verification of the improvements. In the revised manuscript, we will expand these sections to describe the segmentation architecture, the training protocol, the composition of the test set, the number of random seeds, and the statistical significance testing used. This will allow readers to better assess whether the gains are robust. revision: yes

  2. Referee: [Methodology / Dataset Generation] Dataset generation and methodology: no quantitative validation is provided that the automatically generated semantic masks are pixel-accurate on real UAV distributions (e.g., no held-out mask audit, boundary precision metrics, or label-distribution distance between synthetic and expert annotations). Without such checks, the observed gains could stem from the model learning VLM-specific textures or hallucinated boundaries rather than true regeneration patterns, directly undermining the central claim that the synthetic data is “highly complementary.”

    Authors: We acknowledge this limitation in the current manuscript. The central claim relies on the empirical observation that mixing real and synthetic data improves performance on held-out real test images, which we interpret as evidence of complementarity and genuine pattern learning. To address the referee's concern, we will add quantitative comparisons of label distributions (e.g., using Earth Mover's Distance or similar metrics) between the synthetic and real datasets in the revised version. We will also expand the discussion to explicitly consider the possibility of synthetic artifacts and how our evaluation protocol mitigates this. A full pixel-level audit of synthetic masks is not feasible without additional expert annotation effort on the generated images, which we will note as a direction for future work. revision: partial

Circularity Check

0 steps flagged

No significant circularity: purely empirical dataset-augmentation study

full rationale

The paper reports an empirical workflow: a VLM (Nano Banana Pro) is used to synthesize image-mask pairs, these are mixed with newly collected real UAV images, and semantic-segmentation models are trained and evaluated on held-out real data. All headline results (F1 gains of >15 pt overall and up to 30 pt on rare species) are direct experimental measurements against purely supervised baselines. No equations, fitted parameters, uniqueness theorems, or self-citations are invoked as load-bearing steps in any derivation chain. The central claim therefore does not reduce to its own inputs by construction and remains externally falsifiable via the reported train/test splits on real imagery.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the unverified assumption that the chosen vision-language model can produce pixel-accurate semantic masks for complex, heterogeneous forest scenes without introducing biases that would degrade downstream real-world performance.

axioms (1)
  • domain assumption Synthetic images and masks generated by the vision-language model are of high enough fidelity that models trained on them generalize to real UAV imagery.
    This assumption is required for the reported F1 gains to be meaningful on real test data.

pith-pipeline@v0.9.0 · 5664 in / 1297 out tokens · 28811 ms · 2026-05-08T15:04:23.766677+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

99 extracted references · 64 canonical work pages · 7 internal anchors

  1. [1]

    Forestry: An International Journal of Forest Research , volume =

    Achim, Alexis and Moreau, Guillaume and Coops, Nicholas C and Axelson, Jodi N and Barrette, Julie and B. Forestry: An International Journal of Forest Research , volume =. doi:10.1093/forestry/cpab047 , urldate =

    work page doi:10.1093/forestry/cpab047

  2. [2]

    Information Processing in Agriculture , year =

    Tailored refinement of vision-language models for plant instance segmentation , author =. Information Processing in Agriculture , year =

  3. [3]

    Semantic Image Synthesis via Diffusion Models , year=

    Zhou, Wengang and Wang, Weilun and Bao, Jianmin and Chen, Dongdong and Chen, Dong and Yuan, Lu and Li, Houqiang , journal=. Semantic Image Synthesis via Diffusion Models , year=

  4. [4]

    The Cityscapes Dataset for Semantic Urban Scene Understanding , year=

    Cordts, Marius and Omran, Mohamed and Ramos, Sebastian and Rehfeld, Timo and Enzweiler, Markus and Benenson, Rodrigo and Franke, Uwe and Roth, Stefan and Schiele, Bernt , booktitle=. The Cityscapes Dataset for Semantic Urban Scene Understanding , year=

  5. [5]

    2024 , volume =

    Xie, Jiahao and Li, Wei and Li, Xiangtai and Liu, Ziwei and Ong, Yew Soon and Loy, Chen Change , title =. 2024 , volume =. doi:10.1007/s11263-024-02223-3 , journal =

    work page doi:10.1007/s11263-024-02223-3 2024

  6. [6]

    Gen-n-Val: Agentic Image Data Generation and Validation

    Huang, Jing and Fang, I-Sheng and Huang, Tzuhsuan and Wang, Chih-Yu and Chen, Jun-Cheng , title =. 2025 , number =. 2506.04676 , publisher =

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  7. [7]

    Ferreira, Matheus Pinheiro and de Almeida, Danilo Roberti Alves and de Castro Papa, Daniel and Minervino, Jose Belasque and Veras, Hugo Francisco Maia and Formighieri, Andrea and Mesquita, Ricardo Keichi and Straker, Tom and Knight, Christopher and Shimabukuro, Yosio Edemir , journal=

  8. [8]

    Remote Sensing , VOLUME =

    Troles, Jonas and Schmid, Ute and Fan, Wen and Tian, Jiaojiao , TITLE =. Remote Sensing , VOLUME =. 2024 , NUMBER =

    2024

  9. [9]

    2026 , number =

    Duguay, Simon-Olivier and Baudchon, Hugo and Lalibert\'. 2026 , number =. 2602.02426 , publisher =

    work page arXiv 2026

  10. [10]

    Computers and Electronics in Agriculture , volume =

    Towards operational. Computers and Electronics in Agriculture , volume =. 2024 , issn =. doi:https://doi.org/10.1016/j.compag.2024.108785 , url =

    work page doi:10.1016/j.compag.2024.108785 2024

  11. [11]

    Remote Sensing , VOLUME =

    Lv, Lujin and Li, Xuejian and Mao, Fangjie and Zhou, Lv and Xuan, Jie and Zhao, Yinyin and Yu, Jiacong and Song, Meixuan and Huang, Lei and Du, Huaqiang , TITLE =. Remote Sensing , VOLUME =. 2023 , NUMBER =

    2023

  12. [12]

    2024 , volume=

    Wen, Bowen and Yang, Wei and Kautz, Jan and Birchfield, Stan , booktitle=. 2024 , volume=

    2024

  13. [13]

    Semantic understanding of scenes through the

    Zhou, Bolei and Zhao, Hang and Puig, Xavier and Xiao, Tete and Fidler, Sanja and Barriuso, Adela and Torralba, Antonio , journal=. Semantic understanding of scenes through the. 2019 , publisher=

    2019

  14. [14]

    Antropov, Oleg and Molinier, Matthieu and Kuzu, R. Semi-. doi:10.1109/IGARSS52108.2023.10282283 , urldate =

    work page doi:10.1109/igarss52108.2023.10282283 2023

  15. [15]

    Post and Elena A

    Zhenbang Hao and Lili Lin and Christopher J. Post and Elena A. Mikhailova and Kunyong Yu and Huirong Fang and Jian Liu , title =. International Journal of Digital Earth , volume =. 2023 , publisher =

    2023

  16. [16]

    Cubuk, Ekin Dogus and Zoph, Barret and Shlens, Jon and Le, Quoc , booktitle =

  17. [17]

    Leveraging

    Grondin, Vincent and Massicotte, Philippe and Gaha, Mohamed and Pomerleau, Fran. Leveraging. Conference on Robots and Vision (CRV) , year =

  18. [18]

    2501.09008 , publisher =

    Bhat, Aditya and Bose, Rupak and Nwoye, Chinedu Innocent and Padoy, Nicolas , year =. 2501.09008 , publisher =

    work page arXiv

  19. [19]

    doi:10.1038/s41597-026-06665-x , urldate =

    Blaga, Bianca-Cerasela-Zelia and Nedevschi, Sergiu , year = 2026, _month = jan, journal =. doi:10.1038/s41597-026-06665-x , urldate =

    work page doi:10.1038/s41597-026-06665-x 2026

  20. [20]

    2503.19661 , publisher =

    Bose, Rupak and Nwoye, Chinedu Innocent and Bhat, Aditya and Padoy, Nicolas , year =. 2503.19661 , publisher =

    work page arXiv

  21. [21]

    Chen, Beiqi and Wang, Liangjing and Fan, Xijian and Bo, Weihao and Yang, Xubing and Tjahjadi, Tardi , year = 2023, _month = aug, journal =. Semi-. doi:10.3390/rs15164012 , urldate =

    work page doi:10.3390/rs15164012 2023

  22. [22]

    and Kirillov, Alexander and Girdhar, Rohit , booktitle=

    Cheng, Bowen and Misra, Ishan and Schwing, Alexander G. and Kirillov, Alexander and Girdhar, Rohit , booktitle=. Masked-attention Mask Transformer for Universal Image Segmentation , year=

  23. [23]

    Influence of Temperate Forest Autumn Leaf Phenology on Segmentation of Tree Species from

    Cloutier, Myriam and Germain, Micka. Influence of Temperate Forest Autumn Leaf Phenology on Segmentation of Tree Species from. Remote Sensing of Environment , volume =. doi:10.1016/j.rse.2024.114283 , urldate =

    work page doi:10.1016/j.rse.2024.114283 2024

  24. [24]

    and Achim, Alexis and Arp, Paul and Bater, Christopher W

    Coops, Nicholas C. and Achim, Alexis and Arp, Paul and Bater, Christopher W. and Caspersen, John P. and C. Advancing the Application of Remote Sensing for Forest Information Needs in. The Forestry Chronicle , volume =. doi:10.5558/tfc2021-014 , urldate =

    work page doi:10.5558/tfc2021-014

  25. [25]

    Diez, Yago and Kentsch, Sarah and Fukuda, Motohisa and Caceres, Maximo Larry Lopez and Moritake, Koma and Cabezas, Mariano , year = 2021, _month = jul, journal =. Deep. doi:10.3390/rs13142837 , urldate =

    work page doi:10.3390/rs13142837 2021

  26. [26]

    2025 , number =

    Duclos, David-Alexandre and Guimont-Martin, William and Jeanson, Gabriel and Larochelle-Tremblay, Arthur and Defosse, Théo and Moore, Frédéric and Nolet, Philippe and Pomerleau, François and Giguère, Philippe , title =. 2025 , number =. 2510.09458 , publisher =

    work page arXiv 2025

  27. [27]

    Lacking Data?

    Durand, Simon and Foucher, Samuel and Delplanque, Alexandre and Taillon, Jo. Lacking Data?. Remote Sensing in Ecology and Conservation , pages =. doi:10.1002/rse2.70056 , urldate =

    work page doi:10.1002/rse2.70056

  28. [28]

    Cut, Paste and Learn: Surprisingly Easy Synthesis for Instance Detection , year=

    Dwibedi, Debidatta and Misra, Ishan and Hebert, Martial , booktitle=. Cut, Paste and Learn: Surprisingly Easy Synthesis for Instance Detection , year=

  29. [29]

    Forestry: An International Journal of Forest Research , volume =

    Forest Practitioners' Requirements for Remote Sensing-Based Canopy Height, Wood-Volume, Tree Species, and Disturbance Products , author =. Forestry: An International Journal of Forest Research , volume =. doi:10.1093/forestry/cpae021 , urldate =

    work page doi:10.1093/forestry/cpae021

  30. [30]

    Remote Sensing in Forestry: Current Challenges, Considerations and Directions , shorttitle =

    Fassnacht, Fabian Ewald and White, Joanne C and Wulder, Michael A and N. Remote Sensing in Forestry: Current Challenges, Considerations and Directions , shorttitle =. Forestry: An International Journal of Forest Research , volume =. doi:10.1093/forestry/cpad024 , urldate =

    work page doi:10.1093/forestry/cpad024

  31. [31]

    Automated

    Fromm, Michael and Schubert, Matthias and Castilla, Guillermo and Linke, Julia and McDermid, Greg , year = 2019, _month = nov, journal =. Automated. doi:10.3390/rs11212585 , urldate =

    work page doi:10.3390/rs11212585 2019

  32. [32]

    2022 , number =

    Grondin, Vincent and Pomerleau, François and Giguère, Philippe , title =. 2022 , number =. 2210.04104 , publisher =

    work page arXiv 2022

  33. [33]

    Ge, Shaojia and Gu, Hong and Su, Weimin and L. A. IEEE Geoscience and Remote Sensing Letters , volume =. doi:10.1109/LGRS.2023.3281526 , urldate =

    work page doi:10.1109/lgrs.2023.3281526 2023

  34. [34]

    and Coops, Nicholas C

    Goodbody, Tristan R.H. and Coops, Nicholas C. and Marshall, Peter L. and Tompalski, Piotr and Crawford, Patrick , year = 2017, _month = jan, journal =. Unmanned Aerial Systems for Precision Forest Inventory Purposes:. doi:10.5558/tfc2017-012 , urldate =

    work page doi:10.5558/tfc2017-012 2017

  35. [35]

    Generative

    Goodfellow, Ian and. Generative. Advances in

  36. [36]

    Forestry: An International Journal of Forest Research , volume =

    Tree Detection and Diameter Estimation Based on Deep Learning , author =. Forestry: An International Journal of Forest Research , volume =. doi:10.1093/forestry/cpac043 , urldate =

    work page doi:10.1093/forestry/cpac043

  37. [37]

    Proceedings of the 34th International Conference on Neural Information Processing Systems , articleno =

    Ho, Jonathan and Jain, Ajay and Abbeel, Pieter , title =. Proceedings of the 34th International Conference on Neural Information Processing Systems , articleno =. 2020 , isbn =

    2020

  38. [38]

    Huang, Yunmei and Ou, Botong and Meng, Kexin and Yang, Baijian and Carpenter, Joshua and Jung, Jinha and Fei, Songlin , year = 2024, _month = oct, journal =. Tree. doi:10.3390/rs16203836 , urldate =

    work page doi:10.3390/rs16203836 2024

  39. [39]

    Segment Anything

    Kirillov, Alexander and Mintun, Eric and Ravi, Nikhila and Mao, Hanzi and Rolland, Chloe and Gustafson, Laura and Xiao, Tete and Whitehead, Spencer and Berg, Alexander C. and Lo, Wan-Yen and Dollár, Piotr and Girshick, Ross , title =. 2023 , number =. 2304.02643 , publisher =

    work page internal anchor Pith review arXiv 2023

  40. [40]

    Lai, Xin and Tian, Zhuotao and Chen, Yukang and Li, Yanwei and Yuan, Yuhui and Liu, Shu and Jia, Jiaya , year = 2024, _month = jun, pages =. 2024. doi:10.1109/CVPR52733.2024.00915 , urldate =

    work page doi:10.1109/cvpr52733.2024.00915 2024

  41. [41]

    LeCun, Y

    Deep Learning , author =. Nature , volume =. doi:10.1038/nature14539 , urldate =

    work page doi:10.1038/nature14539

  42. [42]

    Classification of Tree Species and Stock Volume Estimation in Ground Forest Images Using

    Liu, Jiazheng and Wang, Xuefeng and Wang, Tian , year = 2019, _month = nov, journal =. Classification of Tree Species and Stock Volume Estimation in Ground Forest Images Using. doi:10.1016/j.compag.2019.105012 , urldate =

    work page doi:10.1016/j.compag.2019.105012 2019

  43. [43]

    2008 , address =

    Sustainable Management in the Boreal Forest: A Real Response to Environmental Challenges , institution =. 2008 , address =

    2008

  44. [44]

    Pattern to Process, Research to Practice: Remote Sensing of Plant Invasions , shorttitle =

    M. Pattern to Process, Research to Practice: Remote Sensing of Plant Invasions , shorttitle =. Biological Invasions , volume =. doi:10.1007/s10530-023-03150-z , urldate =

    work page doi:10.1007/s10530-023-03150-z

  45. [45]

    Using Citizen Science Data as Pre-Training for Semantic Segmentation of High-Resolution

    Nasiri, Kamyar and. Using Citizen Science Data as Pre-Training for Semantic Segmentation of High-Resolution. Forests , volume =. doi:10.3390/f16040616 , urldate =

    work page doi:10.3390/f16040616

  46. [46]

    DINOv2: Learning Robust Visual Features without Supervision

    Oquab, Maxime and Darcet, Timothée and Moutakanni, Théo and Vo, Huy and Szafraniec, Marc and Khalidov, Vasil and Fernandez, Pierre and Haziza, Daniel and Massa, Francisco and El-Nouby, Alaaeldin and Assran, Mahmoud and Ballas, Nicolas and Galuba, Wojciech and Howes, Russell and Huang, Po-Yao and Li, Shang-Wen and Misra, Ishan and Rabbat, Michael and Sharm...

    work page internal anchor Pith review arXiv 2024

  47. [47]

    OpenForest: a data catalog for machine learning in forest monitoring , volume=

    Ouaknine, Arthur and Kattenborn, Teja and Laliberté, Etienne and Rolnick, David , year=. OpenForest: a data catalog for machine learning in forest monitoring , volume=. doi:10.1017/eds.2024.53 , journal=

    work page doi:10.1017/eds.2024.53 2024

  48. [48]

    Hierarchical Text-Conditional Image Generation with CLIP Latents

    Ramesh, Aditya and Dhariwal, Prafulla and Nichol, Alex and Chu, Casey and Chen, Mark , title =. 2022 , number =. 2204.06125 , publisher =

    work page internal anchor Pith review arXiv 2022

  49. [49]

    Environmental Data Science , volume =

    Tree Semantic Segmentation from Aerial Image Time Series , author =. Environmental Data Science , volume =. doi:10.1017/eds.2025.10013 , urldate =

    work page doi:10.1017/eds.2025.10013 2025

  50. [50]

    U-Net: Convolutional Networks for Biomedical Image Segmentation

    Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas , title =. 2015 , number =. 1505.04597 , publisher =

    work page internal anchor Pith review arXiv 2015

  51. [51]

    Effects of Intensified Silviculture on Timber Production and Its Economic Profitability in Boreal

    Routa, J and Kilpel. Effects of Intensified Silviculture on Timber Production and Its Economic Profitability in Boreal. Forestry: An International Journal of Forest Research , volume =. doi:10.1093/forestry/cpz043 , urldate =

    work page doi:10.1093/forestry/cpz043

  52. [52]

    and Cogswell, Michael and Das, Abhishek and Vedantam, Ramakrishna and Parikh, Devi and Batra, Dhruv , booktitle=

    Selvaraju, Ramprasaath R. and Cogswell, Michael and Das, Abhishek and Vedantam, Ramakrishna and Parikh, Devi and Batra, Dhruv , booktitle=. 2017 , volume=

    2017

  53. [53]

    Biogeosciences , volume =

    Automated Mask Generation in Citizen Science Smartphone Photos and Their Value for Mapping Plant Species in Drone Imagery , author =. Biogeosciences , volume =. doi:10.5194/bg-22-6545-2025 , urldate =

    work page doi:10.5194/bg-22-6545-2025 2025

  54. [54]

    Domain-adversarial training of neural networks , year =

    Ganin, Yaroslav and Ustinova, Evgeniya and Ajakan, Hana and Germain, Pascal and Larochelle, Hugo and Laviolette, Fran. Domain-adversarial training of neural networks , year =. Journal of Machine Learning Research , month = jan, pages =

  55. [55]

    From Simple Labels to Semantic Image Segmentation: Leveraging Citizen Science Plant Photographs for Tree Species Mapping in Drone Imagery , shorttitle =

    Soltani, Salim and Ferlian, Olga and Eisenhauer, Nico and Feilhauer, Hannes and Kattenborn, Teja , year = 2024, _month = jun, journal =. From Simple Labels to Semantic Image Segmentation: Leveraging Citizen Science Plant Photographs for Tree Species Mapping in Drone Imagery , shorttitle =. doi:10.5194/bg-21-2909-2024 , urldate =

    work page doi:10.5194/bg-21-2909-2024 2024

  56. [56]

    Transfer Learning from Citizen Science Photographs Enables Plant Species Identification in

    Soltani, Salim and Feilhauer, Hannes and Duker, Robbert and Kattenborn, Teja , year = 2022, _month = aug, journal =. Transfer Learning from Citizen Science Photographs Enables Plant Species Identification in. doi:10.1016/j.ophoto.2022.100016 , urldate =

    work page doi:10.1016/j.ophoto.2022.100016 2022

  57. [57]

    The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences , volume =

    Comparison of Manual and Semi-Automated Synthetic Training Data Creation for Individual Tree Crown Delineation , author =. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences , volume =. doi:10.5194/isprs-archives-XLVIII-1-W6-2025-227-2025 , urldate =

    work page doi:10.5194/isprs-archives-xlviii-1-w6-2025-227-2025 2025

  58. [58]

    International Journal of Applied Earth Observation and Geoinformation , volume =

    High-Resolution Mapping of Tree Species and Associated Uncertainty by Combining Aerial Remote Sensing Data and Convolutional Neural Networks Ensemble , author =. International Journal of Applied Earth Observation and Geoinformation , volume =. doi:10.1016/j.jag.2024.103960 , urldate =

    work page doi:10.1016/j.jag.2024.103960 2024

  59. [59]

    and White, Joanne C

    Tompalski, Piotr and Coops, Nicholas C. and White, Joanne C. and Goodbody, Tristan R.H. and Hennigar, Chris R. and Wulder, Michael A. and Socha, Jaros. Estimating. Current Forestry Reports , volume =. doi:10.1007/s40725-021-00135-w , urldate =

    work page doi:10.1007/s40725-021-00135-w

  60. [60]

    A survey on semi -supervised learning

    A Survey on Semi-Supervised Learning , author =. Machine Learning , volume =. doi:10.1007/s10994-019-05855-6 , urldate =

    work page doi:10.1007/s10994-019-05855-6

  61. [61]

    Benchmarking Representation Learning for Natural World Image Collections , year=

    Van Horn, Grant and Cole, Elijah and Beery, Sara and Wilber, Kimberly and Belongie, Serge and MacAodha, Oisin , booktitle=. Benchmarking Representation Learning for Natural World Image Collections , year=

  62. [62]

    Automation in Construction , volume =

    Controllable Reference-Based Semantic Crack-Image Generation Using Diffusion Model for Intelligent Infrastructure Inspection , author =. Automation in Construction , volume =. doi:10.1016/j.autcon.2025.106759 , urldate =

    work page doi:10.1016/j.autcon.2025.106759 2025

  63. [63]

    doi:10.48550/arXiv.2310.15160 , urldate =

    Yang, Lihe and Xu, Xiaogang and Kang, Bingyi and Shi, Yinghuan and Zhao, Hengshuang , year = 2023, _month = oct, number =. doi:10.48550/arXiv.2310.15160 , urldate =. arXiv , keywords =:2310.15160 , primaryclass =

    work page doi:10.48550/arxiv.2310.15160 2023

  64. [64]

    doi:10.1109/TPAMI.2025.3528453 , urldate =

    Yang, Lihe and Zhao, Zhen and Zhao, Hengshuang , year = 2025, _month = apr, journal =. doi:10.1109/TPAMI.2025.3528453 , urldate =

    work page doi:10.1109/tpami.2025.3528453 2025

  65. [65]

    Computer

    Zhang, Zheng and Ma, Yeyao and Zhang, Enming and Bai, Xiang , editor =. Computer. doi:10.1007/978-3-031-72754-2_5 , urldate =

    work page doi:10.1007/978-3-031-72754-2_5

  66. [66]

    Semi-Supervised

    Zhang, Bin and Zhang, Yongjun and Li, Yansheng and Wan, Yi and Guo, Haoyu and Zheng, Zhi and Yang, Kun , year = 2023, journal =. Semi-Supervised. doi:10.1109/JSTARS.2022.3203750 , urldate =

    work page doi:10.1109/jstars.2022.3203750 2023

  67. [67]

    Beyond Supervised Learning in Remote Sensing: A Systematic Review of Deep Learning Approaches , year=

    Hosseiny, Benyamin and Mahdianpari, Masoud and Hemati, Mohammadali and Radman, Ali and Mohammadimanesh, Fariba and Chanussot, Jocelyn , journal=. Beyond Supervised Learning in Remote Sensing: A Systematic Review of Deep Learning Approaches , year=

  68. [68]

    Explainable Identification and Mapping of Trees Using

    Onishi, Masanori and Ise, Takeshi , year = 2021, _month = jan, journal =. Explainable Identification and Mapping of Trees Using. doi:10.1038/s41598-020-79653-9 , urldate =

    work page doi:10.1038/s41598-020-79653-9 2021

  69. [69]

    Application of

    Chang, Baohua and Li, Fei and Hu, Yuncai and Yin, Hang and Feng, Zhenhua and Zhao, Liang , year = 2025, _month = may, journal =. Application of. doi:10.3389/fpls.2025.1452053 , urldate =

    work page doi:10.3389/fpls.2025.1452053 2025

  70. [70]

    Deep Learning , author=

  71. [71]

    Forestry: An International Journal of Forest Research , volume =

    A Remote Sensing Methodology for Sub-Stand Growth and Yield Projection of Post-Harvest Forest Regeneration , author =. Forestry: An International Journal of Forest Research , volume =. doi:10.1093/forestry/cpae051 , urldate =

    work page doi:10.1093/forestry/cpae051

  72. [72]

    and Coops, Nicholas C

    Chadwick, Andrew J. and Coops, Nicholas C. and Bater, Christopher W. and Martens, Lee A. and White, Barry , year = 2024, _month = jun, journal =. Transferability of a. doi:10.1016/j.srs.2023.100109 , urldate =

    work page doi:10.1016/j.srs.2023.100109 2024

  73. [73]

    Mapping Forest Tree Species in High Resolution

    Schiefer, Felix and Kattenborn, Teja and Frick, Annett and Frey, Julian and Schall, Peter and Koch, Barbara and Schmidtlein, Sebastian , year = 2020, _month = dec, journal =. Mapping Forest Tree Species in High Resolution. doi:10.1016/j.isprsjprs.2020.10.015 , urldate =

    work page doi:10.1016/j.isprsjprs.2020.10.015 2020

  74. [74]

    and Zhao, Yunpeng and Chen, Yuefeng and Liu, Fei and Feng, Xuping , year = 2025, _month = jul, journal =

    Hou, Bingru and Lin, Chenfeng and Chen, Mengyuan and Gouda, Mostafa M. and Zhao, Yunpeng and Chen, Yuefeng and Liu, Fei and Feng, Xuping , year = 2025, _month = jul, journal =. Integrating. doi:10.3390/rs17152541 , urldate =

    work page doi:10.3390/rs17152541 2025

  75. [75]

    2024 , number =

    Jayasumana, Sadeep and Ramalingam, Srikumar and Veit, Andreas and Glasner, Daniel and Chakrabarti, Ayan and Kumar, Sanjiv , title =. 2024 , number =. 2401.09603 , publisher =

    work page arXiv 2024

  76. [76]

    Proceedings of the 31st International Conference on Neural Information Processing Systems , pages =

    Heusel, Martin and Ramsauer, Hubert and Unterthiner, Thomas and Nessler, Bernhard and Hochreiter, Sepp , title =. Proceedings of the 31st International Conference on Neural Information Processing Systems , pages =. 2017 , isbn =

    2017

  77. [77]

    Embracing limited and imperfect training datasets: opportunities and challenges in plant disease recognition using deep learning , volume =

    Xu, Mingle and Kim, Hyongsuk and Yang, Jucheng and Fuentes, Alvaro and Meng, Yao and Yoon, Sook and Kim, Taehyun and Park, Dong Sun , year =. Embracing limited and imperfect training datasets: opportunities and challenges in plant disease recognition using deep learning , volume =. doi:10.3389/fpls.2023.1225409 , journal =

    work page doi:10.3389/fpls.2023.1225409 2023

  78. [78]

    2023 , url =

    Glenn Jocher and Ayush Chaurasia and Jing Qiu , title =. 2023 , url =

    2023

  79. [79]

    doi:10.1007/978-3-030-75178-4 , urldate =

    Synthetic Data for Deep Learning , author =. doi:10.1007/978-3-030-75178-4 , urldate =

    work page doi:10.1007/978-3-030-75178-4

  80. [80]

    2019 , volume=

    Yun, Sangdoo and Han, Dongyoon and Chun, Sanghyuk and Oh, Seong Joon and Yoo, Youngjoon and Choe, Junsuk , booktitle=. 2019 , volume=

    2019

Showing first 80 references.