Pith. sign in

REVIEW 3 major objections 5 minor 65 references

EcoWikiRS: Learning Ecological Representation of Satellite Images from Weak Supervision with Species Observations and Wikipedia

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read This paper claims that fine-tuning remote sensing vision-language models on aerial images paired with co-located species observations and Wikipedia habitat text, using a weighted contrastive loss, improves zero-shot ecosystem…

desk verdict Solid dataset-plus-loss paper for RS ecology: the empirical gain is real but modest, and the mechanistic story is weaker than the headline. read the letter →

arxiv 2504.19742 v1 pith:D4C7E65X submitted 2025-04-28 cs.CV

Valerie Zermatten , Javiera Castillo-Navarro , Pallavi Jain , Devis Tuia , Diego Marcos This is my paper · ORCID
classification cs.CV
keywords remotesensingvision-languagemodelsweaksupervisioncontrastivelearningspeciesobservationsWikipediahabitattextEUNISecosystemclassificationzero-shot
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

The paper aims to show that the ecological character of a landscape can be read from aerial imagery without hand-labeled ecological maps, by using species observations as the link: where a species was recorded, the image of that 100-meter patch is paired with sentences from the species' Wikipedia habitat description. Because most of those sentences describe the species generally rather than the particular patch, the supervision is weak and noisy; the paper's proposed WINCEL loss is a weighted InfoNCE that lets the model itself concentrate on the few relevant sentences. On zero-shot classification of 25 EUNIS ecosystem types in Switzerland, this raises CLIP's overall accuracy from 14.7% to 30.9% and SkyCLIP's from 19.2% to 30.1%, exceeding standard InfoNCE fine-tuning. A sympathetic reader would care because, if the claim holds, text written by naturalists becomes a scalable source of ecological supervision for overhead imagery.

What carries the argument

The central object is WINCEL, a weighted version of the InfoNCE contrastive loss. For an image $I_n$, the visual encoder produces embedding $V_n$ and the text encoder produces $K$ sentence embeddings $T_{n,k}$; each sentence receives weight $\alpha_{n,k}=\sigma(V_n\cdot T_{n,k}/\tau)$, the softmax of image-sentence similarity over the $K$ sentences, and the text representation is $G_n=\sum_k \alpha_{n,k}T_{n,k}$. The loss is then the InfoNCE loss between $V_n$ and $G_n$. This works as a form of text-space interpolation, analogous to mixup, in which the model's own relevance estimates decide how much each sentence contributes, thereby down-weighting false positives and, by pulling text closer to images, reducing false negatives in the batch.

What would settle it

Take a held-out set of images and inspect the single sentence WINCEL gives the highest weight: if, for most images, that sentence's habitat keywords do not match the ground-truth EUNIS ecosystem type while overall accuracy stays high, the gain is coming from the text set rather than from genuine image-text alignment. A sharper version is to train on images whose paired sentences are swapped across unrelated habitats; if zero-shot accuracy is unchanged, the pairing mechanism is not doing the work.

Watch

Extended reading notes

Core claim

The paper's central claim is that a pretrained vision-language model, fine-tuned on the EcoWikiRS dataset with WINCEL, learns an embedding in which aerial images are aligned with ecologically pertinent text rather than with the full noisy set of Wikipedia sentences attached to a location. The discovery is that a softmax-weighted mixture of the K sentence embeddings, with weights set by the model's current image-text similarity, can extract the ecological signal from weak supervision: the paper starts from the hypothesis that at least one sentence per species article describes the image content, and WINCEL learns to select it during training. The reported evidence is the zero-shot EUNIS ecosystem classification performance, where fine-tuning with WINCEL improves all pretrained backbones tested and beats InfoNCE fine-tuning on three of four backbones.

Load-bearing premise

The load-bearing premise is that for each aerial image, at least one of the Wikipedia sentences attached through co-located species observations actually describes what is visible in that specific 100 by 100 meter image; if most sentences are visually irrelevant, WINCEL's weighting has no correct sentence to select and the learned alignment cannot recover the ecological signal.

Editorial extensions

If this is right

  • Zero-shot ecosystem mapping becomes feasible from high-resolution aerial images alone, since after fine-tuning the model can be prompted with EUNIS habitat names without any supervised training on ecosystem labels.
  • General-purpose CLIP benefits most from this fine-tuning, indicating that ecological semantics can be added to a model that was never pretrained on remote sensing imagery.
  • The type of input text matters: sentences extracted from Wikipedia habitat sections outperform keyword-filtered, random, and species-name texts, so the dataset construction protocol is part of the method's effectiveness.
  • WINCEL also changes what the model judges relevant: fine-tuning shifts top-ranked sentences toward ecologically specific descriptions such as soils, urban noise, and alpine zones, and gives negative scores to irrelevant passages, which supports text-based interpretation and retrieval from aerial images.

Reading between the lines

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

  • If the alignment transfers beyond Switzerland, the same recipe of co-located species observations plus Wikipedia habitat text could produce continent-scale ecological maps from freely available coarser satellite imagery; the paper itself flags coarse-resolution free imagery as a scaling bottleneck, so this is an extension rather than a result.
  • A per-species reliability prior could strengthen the weighting: species with broad habitat ranges, the generalists, are exactly where the at-least-one-relevant-sentence assumption is weakest, and the paper's blackbird examples show the model's margin there is thin.
  • Because WINCEL's weights depend on the model's current similarity estimates, an iterative or self-distilled version might push accuracy further by re-weighting with an improved encoder; the paper evaluates a single pass.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 introduces EcoWikiRS, a dataset pairing 91,801 high-resolution aerial images (100 m × 100 m, 50 cm resolution, Switzerland) with geolocated GBIF species observations and filtered Wikipedia habitat sentences describing the observed species. It proposes WINCEL, a weighted variant of InfoNCE, in which each image is contrasted against a convex combination of the K associated sentence embeddings, with weights computed from the model's own image–sentence similarities. The authors fine-tune four pretrained vision–language models (CLIP, GeoRSCLIP, RemoteCLIP, SkyCLIP) on this dataset and evaluate zero-shot EUNIS ecosystem type classification. They report that WINCEL outperforms InfoNCE fine-tuning on three of four backbones (CLIP 30.9% vs 25.3% OA; SkyCLIP 30.1% vs 27.1%; GeoRSCLIP 29.5% vs 27.6%), while RemoteCLIP degrades (20.9% vs 22.2%). Ablations compare text source types, fine-tuning strategies, and alternative noise-robust losses (bootstrap, sampling, substring augmentation). Qualitative visualizations show improved cross-modal similarity maps over Switzerland and example sentence-selection behavior.

Significance. If the central claims hold, the paper offers a scalable weak-supervision source for remote sensing vision–language models in ecology, moving beyond land-cover labels toward habitat and ecosystem semantics. The dataset and code are released, which is a concrete community resource. The evaluation on EUNIS, a standardized European habitat taxonomy, is a sensible benchmark, and the comparison across four backbones with reported means and standard deviations strengthens the empirical case. However, two load-bearing aspects need closer scrutiny: (i) WINCEL's sentence-weighting mechanism is a soft self-training loop that lacks external validation, and (ii) the EUNIS downstream task is constructed from plant species communities, the same ecological signal that supplies the training text, so the reported gains may reflect recovery of species-associated text patterns rather than a general 'ecological understanding' of imagery. The paper's contribution is valuable, but the mechanism claim and the generality of the ecological representation need additional evidence before the results can be fully accepted.

major comments (3)
  1. [Section 3.2 and Section 6.1] The WINCEL weighting α_{n,k} = σ(V_n·T_{n,k}/τ) is computed from the model's own image–text similarity and then used to construct the target G_n that the same model is trained to align with. This is a soft self-training loop: the model's prior biases in cross-modal similarity are reinforced, and there is no external relevance signal. The paper's claim that WINCEL 'learn[s] to recognize text passages that are relevant to the images' (Figure 1 caption) is therefore not established by the current experiments, which provide only qualitative examples (Figures 4 and 9). Please add a quantitative validation of the selection mechanism, for example: (a) measure precision/recall or rank correlation of the learned weights against human annotations of sentence relevance on a held-out subset; (b) compare WINCEL against an 'oracle-weighted' variant that uses the true EUNIS class to select the most relevant sentence; or (c) show that WINCEL's improvement over InfoNCE persists when the model is initialized with a different pretrained encoder whose similarity rankings are decorrelated. Without such evidence, the gain could be attributed to soft averaging or regularization rather than to identifying visually grounded habitat text.
  2. [Section 4, hypothesis statement] The EUNIS ground-truth map is constructed from plant species communities (Section 3.2), and the training signal in EcoWikiRS is also derived from species observations (GBIF) and Wikipedia habitat descriptions of those species. The zero-shot EUNIS task thus measures the model's ability to map aerial images to species-derived ecological descriptions—essentially the same semantic axis as the training text. This does not invalidate the positive results, but it weakens the conclusion that the model has learned a general 'ecologically meaningful' representation. To support the generalization claim, please add at least one downstream evaluation whose labels are not directly generated from species occurrences, such as independent land-cover mapping (e.g., from authoritative land-cover products), or prediction of climate/soil attributes from held-out stations. Alternatively, explicitly reframe the contribution as recovering species-habitat associations from imagery, and temper the ecological-understanding language accordingly.
  3. [Section 4] The method's key assumption—'at least one sentence per article is likely to describe the content of the image'—is load-bearing but never measured. If for many images the sentence set J_n contains no visually pertinent sentence (e.g., because the observed species is a generalist, the species is not visible from above, or the image depicts a habitat that does not match the article text), WINCEL's softmax weighting has no correct sentence to select, and the learned alignment cannot recover the ecological signal. The paper should quantify how often a relevant sentence exists, for example by checking whether the top-weighted sentence under a trained WINCEL model matches the EUNIS habitat description of the image on a validation subset, or by sampling a small set of images and having annotators judge whether at least one of the K sentences is visually relevant. This would also help interpret the RemoteCLIP result, where WINCEL underperforms InfoNCE, perhaps because the pretrained encoder's similarity rankings are too poor to provide useful weights.
minor comments (5)
  1. [Section 6.3] There is a typo: 'such as such as “sandy”, “urban”, “acidic”' should be 'such as “sandy”, “urban”, “acidic”'.
  2. [Section 3.1 vs Section 6.4] Section 3.1 describes swissIMAGE as 'openly available', while Section 6.4 states that 'the lack of freely available high-resolution aerial imagery hinders the direct generalization'. Please reconcile these statements; if the imagery is openly available but not freely redistributable, say so explicitly.
  3. [Equation (4)] In Eq. (4), the summation index is written as i in the first factor (σ(V_n·T_{n,k}/τ)) and k in the second factor (T_{n,k}); this should be a single consistent index, presumably k.
  4. [Supplementary, Tables 5 and 7] The supplementary material alternates between the names 'EcoWikiRS' and 'WikiRS' (e.g., Table 5, Table 7). Please standardize the dataset name throughout.
  5. [Section 5] The temperature hyperparameters are set to 0.07 for InfoNCE and 0.15 for WINCEL after separate grid searches. Since temperature directly controls the softmax sharpness in both the contrastive loss and the weighting, it would be informative to report a sensitivity analysis over τ for both methods, or at least to state the ranges searched, to ensure the comparison is not driven by this hyperparameter.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: WINCEL's sentence weighting is a self-referential training objective, not a disguised fit of the external EUNIS evaluation.

full rationale

The paper's central quantitative claim is empirical: fine-tuning pretrained RS-VLMs on EcoWikiRS with WINCEL improves zero-shot EUNIS classification relative to InfoNCE and pretrained baselines. The EUNIS labels are external to the EcoWikiRS training signal and are not used in the WINCEL objective, so no fitted parameter is renamed as a prediction and no evaluation result is forced by construction. The sentence weights alpha_{n,k} = sigma(V_n . T_{n,k} / tau) in Eq. (3) are computed from the very model being trained, making WINCEL a soft self-training or bootstrapping objective; however, the paper explicitly attributes this design to the pretrained model's knowledge and does not present sentence relevance as a first-principles derivation. The qualitative relevance evidence in Figures 4 and 9 is self-scored, which weakens the 'integrates ecological knowledge' claim, but that is a validation gap rather than a circular derivation. Self-citations (e.g., [58], [23], [6]) appear in related-work context and are not load-bearing for the main result. No uniqueness theorem or ansatz is imported from the authors' prior work. The hypothesis that at least one Wikipedia sentence is visually relevant to each image is an unverified assumption; if it fails, the method's mechanism degrades, but failing assumptions are not circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the ecological proxy assumption (species presence implies habitat properties), the relevance assumption (at least one Wikipedia sentence per species describes the visual content), the reliability of EUNIS labels for evaluation, and the quality of GBIF observations. The only freely tuned numbers in the method are the two temperature values and the sentence count K, which are standard hyperparameters but do influence the results.

free parameters (3)
  • InfoNCE temperature tau = 0.07
    Temperature for the contrastive logits; selected by grid search separately for each method (Section 5).
  • WINCEL temperature tau = 0.15
    Temperature for both the sentence weighting softmax and the contrastive loss; selected by grid search (Section 5).
  • Number of sentences per sample K = 15
    Sentences are sampled or padded to K per image; set in Section 5 without a sensitivity study.
assumptions (4)
  • domain assumption Species presence at a location is a proxy for local ecological properties such as land cover, climate, and soil.
    Stated in the abstract and introduction as the key premise for using GBIF observations as supervision.
  • domain assumption For each image, at least one sentence among the K Wikipedia habitat sentences is visually relevant to that image.
    Stated in Section 4 as the hypothesis behind WINCEL. This is the load-bearing premise for the loss to work.
  • domain assumption EUNIS Ecosystem Type Map v3.1 provides reliable ground truth for the downstream evaluation.
    The evaluation uses this map in Section 3.2; the authors note in Section 6.4 that the map may contain erroneous labels.
  • domain assumption GBIF observations after filtering are accurate enough for pairing with images.
    The dataset construction filters by coordinate uncertainty below 100 m and removes incomplete records (Supplementary Section 11.2), but the data remain presence-only and biased, as acknowledged in Section 6.4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EcoWikiRS: Learning Ecological Representation of Satellite Images from Weak Supervision with Species Observations and Wikipedia." pith.science (2026). https://pith.science/paper/D4C7E65X

@misc{pith2026250419742,
  author       = {Pith},
  title        = {Pith review of: EcoWikiRS: Learning Ecological Representation of Satellite Images from Weak Supervision with Species Observations and Wikipedia},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D4C7E65X}},
  note         = {Machine review of arXiv:2504.19742}
}
read the original abstract

The presence of species provides key insights into the ecological properties of a location such as land cover, climatic conditions or even soil properties. We propose a method to predict such ecological properties directly from remote sensing (RS) images by aligning them with species habitat descriptions. We introduce the EcoWikiRS dataset, consisting of high-resolution aerial images, the corresponding geolocated species observations, and, for each species, the textual descriptions of their habitat from Wikipedia. EcoWikiRS offers a scalable way of supervision for RS vision language models (RS-VLMs) for ecology. This is a setting with weak and noisy supervision, where, for instance, some text may describe properties that are specific only to part of the species' niche or is irrelevant to a specific image. We tackle this by proposing WINCEL, a weighted version of the InfoNCE loss. We evaluate our model on the task of ecosystem zero-shot classification by following the habitat definitions from the European Nature Information System (EUNIS). Our results show that our approach helps in understanding RS images in a more ecologically meaningful manner. The code and the dataset are available at https://github.com/eceo-epfl/EcoWikiRS.

Figures

Figures reproduced from arXiv: 2504.19742 by the authors.

Figure 1
Figure 1. EcoWikiRS connects aerial images with local species [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. EcoWikiRS dataset preparation. For each location containing species observation in GBIF, an aerial image and its ecosystem [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visualization of text-image similarity values over the surface of Switzerland with different text prompts as inputs. The maps uses [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparison of top-3 sentences scores given by the pretrained and fine-tuned SkyCLIP model on samples of the WikiRS dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Cross-modal similarities values between sentences from the Wikipedia article of [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: The EUNIS ecosystem type map in Switzerland at level L2. [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 8
Figure 8. Figure 8: Distribution into EUNIS ecosystem types of samples [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: Comparison of top-5 sentences scores given by the pretrained and fine-tuned SkyCLIP model on samples of the WikiRS dataset. [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 48 canonical work pages

  1. [1]

    https://www.inaturalist.org

    iNaturalist. https://www.inaturalist.org . Ac- cessed: 2025-03-01. 2, 3

    work page 2025

  2. [2]

    A simple zero-shot prompt weight- ing technique to improve prompt ensembling in text-image models

    James Urquhart Allingham, Jie Ren, Michael W Dusenberry, Xiuye Gu, Yin Cui, Dustin Tran, Jeremiah Zhe Liu, and Bal- aji Lakshminarayanan. A simple zero-shot prompt weight- ing technique to improve prompt ensembling in text-image models. In International Conference on Machine Learning, pages 547–568. PMLR, 2023. 12

    work page 2023

  3. [3]

    Bird- snap: Large-scale fine-grained visual categorization of birds

    Thomas Berg, Jiongxin Liu, Seung Woo Lee, Michelle L Alexander, David W Jacobs, and Peter N Belhumeur. Bird- snap: Large-scale fine-grained visual categorization of birds. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2011–2018, 2014. 3

    work page 2011

  4. [4]

    An unexpectedly large count of trees in the west african sahara and sahel

    Martin Brandt, Compton J Tucker, Ankit Kariryaa, Kjeld Rasmussen, Christin Abel, Jennifer Small, Jerome Chave, Laura Vang Rasmussen, Pierre Hiernaux, Abdoul Aziz Diouf, et al. An unexpectedly large count of trees in the west african sahara and sahel. Nature, 587(7832):78–82, 2020. 1

    work page 2020

  5. [5]

    Unveiling global species abundance distributions

    Corey T Callaghan, Lu ´ıs Borda-de ´Agua, Roel van Klink, Roberto Rozzi, and Henrique M Pereira. Unveiling global species abundance distributions. Nature ecology & evolu- tion, 7(10):1600–1609, 2023. 3

    work page 2023

  6. [6]

    Prompt-rsvqa: Prompting visual context to a language model for remote sensing visual question answering

    Christel Chappuis, Val ´erie Zermatten, Sylvain Lobry, Bertrand Le Saux, and Devis Tuia. Prompt-rsvqa: Prompting visual context to a language model for remote sensing visual question answering. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 1372–1381, 2022. 1

    work page 2022

  7. [7]

    Data-efficient language-supervised zero-shot learning with self-distillation

    Ruizhe Cheng, Bichen Wu, Peizhao Zhang, Peter Vajda, and Joseph E Gonzalez. Data-efficient language-supervised zero-shot learning with self-distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3119–3124, 2021. 2

    work page 2021

  8. [8]

    Towards natural language-guided drones: Geotext-1652 benchmark with spatial relation matching

    Meng Chu, Zhedong Zheng, Wei Ji, Tingyu Wang, and Tat-Seng Chua. Towards natural language-guided drones: Geotext-1652 benchmark with spatial relation matching. In European Conference on Computer Vision, pages 213–231. Springer, 2024. 2

    work page 2024

Show all 65 references
  1. [9]

    Eunis habitat classification: Expert system, character- istic species combinations and distribution maps of european habitats

    Milan Chytr `y, Lubom´ır Tich`y, Stephan M Hennekens, Ilona Knollov´a, John AM Janssen, John S Rodwell, Tom ´aˇs Pe- terka, Corrado Marcen `o, Flavia Landucci, Ji ˇr´ı Danihelka, et al. Eunis habitat classification: Expert system, character- istic species combinations and dist...

    2020

  2. [10]

    WildSAT: Learning Satellite Image Representations from Wildlife Observations, 2024

    Rangel Daroya, Elijah Cole, Oisin Mac Aodha, Grant Van Horn, and Subhransu Maji. WildSAT: Learning Satellite Image Representations from Wildlife Observations, 2024. arXiv:2412.14428 [cs]. 2, 3

    2024 arXiv

  3. [11]

    Widespread sampling biases in herbaria revealed from large-scale digitization

    Barnabas H Daru, Daniel S Park, Richard B Primack, Charles G Willis, David S Barrington, Timothy JS Whit- feld, Tristram G Seidler, Patrick W Sweeney, David R Fos- ter, Aaron M Ellison, et al. Widespread sampling biases in herbaria revealed from large-scale digitization. New P...

    2018

  4. [12]

    Multispecies detection and identification of african mammals in aerial imagery us- ing convolutional neural networks

    Alexandre Delplanque, Samuel Foucher, Philippe Lejeune, Julie Linchant, and J ´erˆome Th ´eau. Multispecies detection and identification of african mammals in aerial imagery us- ing convolutional neural networks. Remote Sensing in Ecol- ogy and Conservation, 8(2):166–179, 2022. 1

    2022

  5. [13]

    Sat2cap: Map- ping fine-grained textual descriptions from satellite images

    Aayush Dhakal, Adeel Ahmad, Subash Khanal, Srikumar Sastry, Hannah Kerner, and Nathan Jacobs. Sat2cap: Map- ping fine-grained textual descriptions from satellite images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 533–542, 2024. 1, 2

    2024

  6. [14]

    With a Little Help from My Friends: Nearest-Neighbor Contrastive Learning of Vi- sual Representations

    Debidatta Dwibedi, Yusuf Aytar, Jonathan Tompson, Pierre Sermanet, and Andrew Zisserman. With a Little Help from My Friends: Nearest-Neighbor Contrastive Learning of Vi- sual Representations. In 2021 IEEE/CVF International Con- ference on Computer Vision (ICCV), Montreal, QC, Canada,

    2021

  7. [15]

    Learning visual representations via language-guided sam- pling

    Mohamed El Banani, Karan Desai, and Justin Johnson. Learning visual representations via language-guided sam- pling. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 19208–19220,

  8. [16]

    Bias correction in species distribution models: pool- ing survey and collection data for multiple species

    William Fithian, Jane Elith, Trevor Hastie, and David A Keith. Bias correction in species distribution models: pool- ing survey and collection data for multiple species. Methods in ecology and evolution, 6(4):424–438, 2015. 8

    2015

  9. [17]

    Wikipedia corpus

    Wikimedia Foundation. Wikipedia corpus. Wikimedia Foundation, The Free Encyclopedia, 2025. Accessed from https://dumps.wikimedia.org/. 4

    2025

  10. [18]

    SoftCLIP: Softer Cross-Modal Alignment Makes CLIP Stronger

    Yuting Gao, Jinfeng Liu, Zihan Xu, Tong Wu, Enwei Zhang, Ke Li, Jie Yang, Wei Liu, and Xing Sun. SoftCLIP: Softer Cross-Modal Alignment Makes CLIP Stronger. Proceed- ings of the AAAI Conference on Artificial Intelligence , 38 (3):1860–1868, 2024. Number: 3. 2

    2024

  11. [19]

    Gbif occurrence download, doi:10.15468/dl.pr8cws, 2024

    GBIF.Org User. Gbif occurrence download, doi:10.15468/dl.pr8cws, 2024. 4, 13

    2024 doi

  12. [20]

    Pl@ ntnet mobile app

    Herv ´e Go¨eau, Pierre Bonnet, Alexis Joly, Vera Baki´c, Julien Barbe, Itheri Yahiaoui, Souheil Selmi, Jennifer Carr´e, Daniel Barth´el´emy, Nozha Boujemaa, et al. Pl@ ntnet mobile app. In Proceedings of the 21st ACM international conference on Multimedia, pages 423–424, 2013. 2

    2013

  13. [21]

    Combining observational data and language for species range estimation

    Max Hamilton, Christian Lange, Elijah Cole, Alexan- der Shepard, Samuel Heinrich, Oisin Mac Aodha, Grant Van Horn, and Subhransu Maji. Combining observational data and language for species range estimation. Advances in Neural Information Processing Systems, 37:17719–17742,

  14. [22]

    Open- clip, 2021

    Gabriel Ilharco, Mitchell Wortsman, Ross Wightman, Cade Gordon, Nicholas Carlini, Rohan Taori, Achal Dave, Vaishaal Shankar, Hongseok Namkoong, John Miller, Han- naneh Hajishirzi, Ali Farhadi, and Ludwig Schmidt. Open- clip, 2021. 5

    2021

  15. [23]

    Senclip: Enhancing zero-shot land-use mapping for sentinel-2 with ground-level prompt- ing

    Pallavi Jain, Dino Ienco, Roberto Interdonato, Tristan Berchoux, and Diego Marcos. Senclip: Enhancing zero-shot land-use mapping for sentinel-2 with ground-level prompt- ing. In Proceedings of the Winter Conference on Applica- tions of Computer Vision (WACV), pages 5656–5665, ...

    2025

  16. [24]

    Scaling up visual and vision-language representa- tion learning with noisy text supervision

    Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc Le, Yun-Hsuan Sung, Zhen Li, and Tom Duerig. Scaling up visual and vision-language representa- tion learning with noisy text supervision. In International conference on machine learning, pages 4904–4916. PMLR,

  17. [25]

    Rapidly mapping fire effects on biodiversity at a large-scale using citizen science

    Casey Kirchhoff, Corey T Callaghan, David A Keith, Dony Indiarto, Guy Taseski, Mark KJ Ooi, Tom D Le Breton, Thomas Mesaglio, Richard T Kingsford, and William K Cornwell. Rapidly mapping fire effects on biodiversity at a large-scale using citizen science. Science of the Total ...

    2021

  18. [26]

    A high-resolution canopy height model of the earth

    Nico Lang, Walter Jetz, Konrad Schindler, and Jan Dirk Wegner. A high-resolution canopy height model of the earth. Nature Ecology & Evolution, 7(11):1778–1789, 2023. 1

    2023

  19. [27]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In Interna- tional conference on machine learning, pages 12888–12900. PMLR, 2022. 2

    2022

  20. [28]

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In In- ternational conference on machine learning , pages 19730– 19742. PMLR, 2023. 2

    2023

  21. [29]

    Selective-supervised contrastive learning with noisy labels

    Shikun Li, Xiaobo Xia, Shiming Ge, and Tongliang Liu. Selective-supervised contrastive learning with noisy labels. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 316–325, 2022. 2, 4

    2022

  22. [30]

    Su- pervision Exists Everywhere: A Data Efficient Contrastive Language-Image Pre-training Paradigm

    Yangguang Li, Feng Liang, Lichen Zhao, Yufeng Cui, Wanli Ouyang, Jing Shao, Fengwei Yu, and Junjie Yan. Su- pervision Exists Everywhere: A Data Efficient Contrastive Language-Image Pre-training Paradigm. In International Conference on Learning Representations, 2022. 2

    2022

  23. [31]

    Re- moteclip: A vision language foundation model for remote sensing

    Fan Liu, Delong Chen, Zhangqingyun Guan, Xiaocong Zhou, Jiale Zhu, Qiaolin Ye, Liyong Fu, and Jun Zhou. Re- moteclip: A vision language foundation model for remote sensing. IEEE Transactions on Geoscience and Remote Sensing, 2024. 1, 2, 5

    2024

  24. [32]

    Decoupled weight de- cay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight de- cay regularization. In International Conference on Learning Representations, 2017. 5

    2017

  25. [33]

    Presence- only geographical priors for fine-grained image classifica- tion

    Oisin Mac Aodha, Elijah Cole, and Pietro Perona. Presence- only geographical priors for fine-grained image classifica- tion. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 9596–9606, 2019. 3

    2019

  26. [34]

    Ro- bust audio-visual instance discrimination

    Pedro Morgado, Ishan Misra, and Nuno Vasconcelos. Ro- bust audio-visual instance discrimination. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12934–12945, 2021. 2, 4

    2021

  27. [35]

    Slip: Self-supervision meets language-image pre- training

    Norman Mu, Alexander Kirillov, David Wagner, and Sain- ing Xie. Slip: Self-supervision meets language-image pre- training. In European conference on computer vision, pages 529–544. Springer, 2022. 2

    2022

  28. [36]

    Lhrs-bot: Empowering remote sensing with vgi-enhanced large multimodal language model

    Dilxat Muhtar, Zhenshi Li, Feng Gu, Xueliang Zhang, and Pengfeng Xiao. Lhrs-bot: Empowering remote sensing with vgi-enhanced large multimodal language model. In European Conference on Computer Vision, pages 440–457. Springer, 2024. 2

    2024

  29. [37]

    Repre- sentation learning with contrastive predictive coding

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Repre- sentation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748, 2018. 2, 4

    2018 arXiv

  30. [38]

    ZEST: Zero-shot Learning from Text Descrip- tions using Textual Similarity and Visual Summarization

    Tzuf Paz-Argaman, Reut Tsarfaty, Gal Chechik, and Yu- val Atzmon. ZEST: Zero-shot Learning from Text Descrip- tions using Textual Similarity and Visual Summarization. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 569–579, Online, 2020. Associa...

    2020

  31. [39]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  32. [40]

    Train- ing deep neural networks on noisy labels with bootstrapping

    Scott Reed, Honglak Lee, Dragomir Anguelov, Christian Szegedy, Dumitru Erhan, and Andrew Rabinovich. Train- ing deep neural networks on noisy labels with bootstrapping. arXiv preprint arXiv:1412.6596, 2014. 2, 6

    2014 arXiv

  33. [41]

    Machine- to-machine visual dialoguing with chatgpt for enriched tex- tual image description

    Riccardo Ricci, Yakoub Bazi, and Farid Melgani. Machine- to-machine visual dialoguing with chatgpt for enriched tex- tual image description. Remote Sensing, 16(3):441, 2024. 1, 2

    2024

  34. [42]

    Beautiful soup documentation

    Leonard Richardson. Beautiful soup documentation. April,

  35. [43]

    The gbif integrated publishing toolkit: facilitating the efficient publishing of biodiversity data on the internet

    Tim Robertson, Markus D ¨oring, Robert Guralnick, David Bloom, John Wieczorek, Kyle Braak, Javier Otegui, Laura Russell, and Peter Desmet. The gbif integrated publishing toolkit: facilitating the efficient publishing of biodiversity data on the internet. PloS one, 9(8):e102623...

    2014

  36. [44]

    Im- proved Zero-Shot Classification by Adapting VLMs with Text Descriptions

    Oindrila Saha, Grant Van Horn, and Subhransu Maji. Im- proved Zero-Shot Classification by Adapting VLMs with Text Descriptions. In 2024 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 17542– 17552, Seattle, W A, USA, 2024. IEEE. 3

    2024

  37. [45]

    TaxaBind: A Unified Embedding Space for Ecological Applications, 2024

    Srikumar Sastry, Subash Khanal, Aayush Dhakal, Adeel Ah- mad, and Nathan Jacobs. TaxaBind: A Unified Embedding Space for Ecological Applications, 2024. arXiv:2411.00683 [cs]. 3

    2024 arXiv

  38. [46]

    Laion-5b: An open large-scale dataset for training next generation image-text models

    Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Worts- man, et al. Laion-5b: An open large-scale dataset for training next generation image-text models. Advances in neural in- f...

    2022

  39. [47]

    Geopixel: Pixel grounding large multimodal model in remote sensing

    Akashah Shabbir, Mohammed Zumri, Mohammed Ben- namoun, Fahad S Khan, and Salman Khan. Geopixel: Pixel grounding large multimodal model in remote sensing. arXiv preprint arXiv:2501.13925, 2025. 2

    2025 arXiv

  40. [48]

    The ebird enterprise: An integrated approach to development and application of citizen science

    Brian L Sullivan, Jocelyn L Aycrigg, Jessie H Barry, Rick E Bonney, Nicholas Bruns, Caren B Cooper, Theo Damoulas, Andr´e A Dhondt, Tom Dietterich, Andrew Farnsworth, et al. The ebird enterprise: An integrated approach to development and application of citizen science. Biologi...

    2014

  41. [49]

    Learning to Interpret Satellite Images using Wikipedia.Proceedings of the Twenty-Eighth International Joint Conference on Artifi- cial Intelligence, 2019

    Burak Uzkent, Evan Sheehan, Chenlin Meng, Zhongyi Tang, Marshall Burke, David Lobell, and Stefano Ermon. Learning to Interpret Satellite Images using Wikipedia.Proceedings of the Twenty-Eighth International Joint Conference on Artifi- cial Intelligence, 2019. 2

    2019

  42. [50]

    Regional uniqueness of tree species compo- sition and response to forest loss and climate change

    Nina van Tiel, Fabian Fopp, Philipp Brun, Johan van den Hoogen, Dirk Nikolaus Karger, Cecilia M Casadei, Lisha Lyu, Devis Tuia, Niklaus E Zimmermann, Thomas W Crowther, et al. Regional uniqueness of tree species compo- sition and response to forest loss and climate change. Nat...

    2024

  43. [51]

    inaturalist in- sights illuminate covid-19 effects on large mammals in urban centers

    Reut Vardi, Oded Berger-Tal, and Uri Roll. inaturalist in- sights illuminate covid-19 effects on large mammals in urban centers. Biological conservation, 254:108953, 2021. 3

    2021

  44. [52]

    SkyScript: A Large and Seman- tically Diverse Vision-Language Dataset for Remote Sens- ing

    Zhecheng Wang, Rajanie Prabha, Tianyuan Huang, Jiajun Wu, and Ram Rajagopal. SkyScript: A Large and Seman- tically Diverse Vision-Language Dataset for Remote Sens- ing. Proceedings of the AAAI Conference on Artificial Intel- ligence, 38(6):5805–5813, 2024. Number: 6. 1, 2, 5

    2024

  45. [53]

    Instructiongpt-4: A 200-instruction paradigm for fine-tuning minigpt-4

    Lai Wei, Zihao Jiang, Weiran Huang, and Lichao Sun. Instructiongpt-4: A 200-instruction paradigm for fine-tuning minigpt-4. arXiv preprint arXiv:2308.12067, 2023. 2

    2023 arXiv

  46. [54]

    Ecosystem type map v3

    Michael Weiss and GJTP Banko. Ecosystem type map v3. 1–terrestrial and marine ecosystems. European Topic Centre on Biological Diversity report to the European Environment Agency (EEA), page 79, 2018. 4

    2018

  47. [55]

    Citi- zen science plant observations encode global trait patterns

    Sophie Wolf, Miguel D Mahecha, Francesco Maria Saba- tini, Christian Wirth, Helge Bruelheide, Jens Kattge, ´Alvaro Moreno Mart´ınez, Karin Mora, and Teja Kattenborn. Citi- zen science plant observations encode global trait patterns. Nature ecology & evolution, 6(12):1850–1859, 2022. 3

    2022

  48. [56]

    Deep learning enables satellite-based monitoring of large popula- tions of terrestrial mammals across heterogeneous landscape

    Zijing Wu, Ce Zhang, Xiaowei Gu, Isla Duporge, Lacey F Hughey, Jared A Stabach, Andrew K Skidmore, J Grant C Hopcraft, Stephen J Lee, Peter M Atkinson, et al. Deep learning enables satellite-based monitoring of large popula- tions of terrestrial mammals across heterogeneous la...

    2023

  49. [57]

    Alip: Adaptive language-image pre-training with synthetic cap- tion

    Kaicheng Yang, Jiankang Deng, Xiang An, Jiawei Li, Ziy- ong Feng, Jia Guo, Jing Yang, and Tongliang Liu. Alip: Adaptive language-image pre-training with synthetic cap- tion. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 2922–2931, 2023. 2

    2023

  50. [58]

    Learning transferable land cover seman- tics for open vocabulary interactions with remote sensing im- ages

    Val ´erie Zermatten, Javiera Castillo-Navarro, Diego Marcos, and Devis Tuia. Learning transferable land cover seman- tics for open vocabulary interactions with remote sensing im- ages. ISPRS Journal of Photogrammetry and Remote Sens- ing, 220:621–636, 2025. 1

    2025

  51. [59]

    mixup: Beyond empirical risk minimiza- tion

    Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. mixup: Beyond empirical risk minimiza- tion. In International Conference on Learning Representa- tions, 2018. 5

    2018

  52. [60]

    RS5M and GeoRSCLIP: A Large Scale Vision-Language Dataset and A Large Vision-Language Model for Remote Sensing

    Zilun Zhang, Tiancheng Zhao, Yulong Guo, and Jianwei Yin. RS5M and GeoRSCLIP: A Large Scale Vision-Language Dataset and A Large Vision-Language Model for Remote Sensing. IEEE Transactions on Geoscience and Remote Sensing, pages 1–1, 2024. arXiv:2306.11300 [cs]. 1, 2, 5

    2024 arXiv

  53. [61]

    No one-size-fits-all solution to clean gbif

    Alexander Zizka, Fernanda Antunes Carvalho, Alice Cal- vente, Mabel Rocio Baez-Lizarazo, Andressa Cabral, J´essica Fernanda Ramos Coelho, Matheus Colli-Silva, Mar- iana Ramos Fantinati, Moabe F Fernandes, Thais Ferreira- Ara´ujo, et al. No one-size-fits-all solution to clean g...

    2020

  54. [62]

    Table 4 compares the performance of SkyCLIP under vari- ous fine-tuning configurations with WINCEL

    Study of fine-tuning strategies. Table 4 compares the performance of SkyCLIP under vari- ous fine-tuning configurations with WINCEL. The text en- coder remains frozen, while different layers of the visual en- coder undergo fine-tuning: the final transformer block, the last pro...

  55. [63]

    Thus, sen- tences describing their habitat are likely to be uninformative or irrelevant when paired with a given image

    Additional visual results As discussed in Section 4, generalist species, such as the common blackbird, can live in several habitats. Thus, sen- tences describing their habitat are likely to be uninformative or irrelevant when paired with a given image. To demon- strate that ou...

  56. [64]

    To select satisfying prompts for the EUNIS ecosystem classification task, we compared manually de- signed prompt templates

    Prompting for zero-shot classification Due to the presence of spurious concept biases, several works highlighted the importance of prompt engineer- ing [2, 39] for attaining high zero-shot classification per- formances. To select satisfying prompts for the EUNIS ecosystem clas...

  57. [65]

    Speciesbox

    Dataset construction and statistics 11.1. Wikipedia articles parsing Extraction of Wikipedia articles. We downloaded a dump of all of Wikipedia through Wikimedia at dumps . wikimedia . org. We processed the dump with the BeautifulSoup [42] and mwparserfromhell python packages....

    1950

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.