mirror of
https://github.com/zoo-apps/papers.git
synced 2026-10-30 14:56:54 +00:00
zeekayandHanzo Dev
6658236499
Leadership and team cards, board and founding prose, an AI persona modelled on a named individual, contributor and partner entries, and personal correspondence addresses all named one person. The organisation carries them now. Other people keep their names. Where a card, a board list or a founding line named someone else alongside, they stay exactly as they were; only the one entry was removed, and prose that listed two names now reads with one. Sample and fixture data uses the placeholder name these tests already use elsewhere, so the IAM username defect the notes describe still reads as written. The zoo/yapp MIT notice names the foundation that publishes it. Co-authored-by: Hanzo Dev <[email protected]>
193 lines
8.4 KiB
TeX
193 lines
8.4 KiB
TeX
\section{Conclusion}
|
|
|
|
\label{sec:conclusion}
|
|
% ══════════════════════════════════════════════════════════════════════════════
|
|
|
|
We have presented TaxoNet, a taxonomy-aware few-shot visual recognition
|
|
system for biodiversity monitoring. By incorporating biological
|
|
classification structure into meta-learning through taxonomic prototypical
|
|
networks and rank-attention fusion, TaxoNet achieves state-of-the-art
|
|
few-shot accuracy across four major taxonomic groups. Integration with
|
|
geographic and phenological priors further boosts practical field
|
|
identification to 94.7\% accuracy. Through Zoo Labs Foundation's
|
|
commitment to open science, TaxoNet's models, benchmarks, and deployment
|
|
infrastructure are freely available to the conservation community.
|
|
|
|
% ══════════════════════════════════════════════════════════════════════════════
|
|
% References
|
|
% ══════════════════════════════════════════════════════════════════════════════
|
|
|
|
\bibliographystyle{plainnat}
|
|
|
|
\begin{thebibliography}{33}
|
|
|
|
\bibitem[Akata et~al.(2015)]{akata2015label}
|
|
Akata, Z., Reed, S., Walter, D., Lee, H., and Schiele, B.
|
|
\newblock Evaluation of output embeddings for fine-grained image classification.
|
|
\newblock \emph{CVPR}, pp. 2927--2936, 2015.
|
|
|
|
\bibitem[Beery et~al.(2021)]{beery2021iwildcam}
|
|
Beery, S., Cole, E., Parker, J., Perona, P., and Winner, K.
|
|
\newblock Species classification in camera traps: The iWildCam 2021 challenge.
|
|
\newblock \emph{FGVC Workshop at CVPR}, 2021.
|
|
|
|
\bibitem[Bengio et~al.(2009)]{bengio2009curriculum}
|
|
Bengio, Y., Louradour, J., Collobert, R., and Weston, J.
|
|
\newblock Curriculum learning.
|
|
\newblock \emph{ICML}, pp. 41--48, 2009.
|
|
|
|
\bibitem[Brown et~al.(2020)]{brown2020language}
|
|
Brown, T., Mann, B., Ryder, N., et~al.
|
|
\newblock Language models are few-shot learners.
|
|
\newblock \emph{NeurIPS}, 33:1877--1901, 2020.
|
|
|
|
\bibitem[Chang et~al.(2021)]{chang2021your}
|
|
Chang, D., Pang, K., Zheng, Y., et~al.
|
|
\newblock Your ``flamingo'' is my ``bird'': Fine-grained, or not.
|
|
\newblock \emph{CVPR}, pp. 11476--11485, 2021.
|
|
|
|
\bibitem[Christin et~al.(2019)]{christin2019applications}
|
|
Christin, S., Hervet, {\'E}., and Lecomte, N.
|
|
\newblock Applications for deep learning in ecology.
|
|
\newblock \emph{Methods in Ecology and Evolution}, 10(10):1632--1644, 2019.
|
|
|
|
\bibitem[Costello et~al.(2013)]{costello2013can}
|
|
Costello, M.J., May, R.M., and Stork, N.E.
|
|
\newblock Can we name Earth's species before they go extinct?
|
|
\newblock \emph{Science}, 339(6118):413--416, 2013.
|
|
|
|
\bibitem[Deng et~al.(2014)]{deng2014large}
|
|
Deng, J., Ding, N., Jia, Y., et~al.
|
|
\newblock Large-scale object classification using label relation graphs.
|
|
\newblock \emph{ECCV}, pp. 48--64, 2014.
|
|
|
|
\bibitem[Dosovitskiy et~al.(2021)]{dosovitskiy2021image}
|
|
Dosovitskiy, A., Beyer, L., Kolesnikov, A., et~al.
|
|
\newblock An image is worth 16x16 words: Transformers for image recognition.
|
|
\newblock \emph{ICLR}, 2021.
|
|
|
|
\bibitem[Finn et~al.(2017)]{finn2017model}
|
|
Finn, C., Abbeel, P., and Levine, S.
|
|
\newblock Model-agnostic meta-learning for fast adaptation of deep networks.
|
|
\newblock \emph{ICML}, pp. 1126--1135, 2017.
|
|
|
|
\bibitem[Go{\"e}au et~al.(2022)]{goeau2022overview}
|
|
Go{\"e}au, H., Bonnet, P., and Joly, A.
|
|
\newblock Overview of PlantCLEF 2022: Image-based plant identification.
|
|
\newblock \emph{CLEF}, 2022.
|
|
|
|
\bibitem[Guo et~al.(2017)]{guo2017calibration}
|
|
Guo, C., Pleiss, G., Sun, Y., and Weinberger, K.Q.
|
|
\newblock On calibration of modern neural networks.
|
|
\newblock \emph{ICML}, pp. 1321--1330, 2017.
|
|
|
|
\bibitem[He et~al.(2022)]{he2022transfg}
|
|
He, J., Chen, J.N., Liu, S., et~al.
|
|
\newblock TransFG: A transformer architecture for fine-grained recognition.
|
|
\newblock \emph{AAAI}, pp. 852--860, 2022.
|
|
|
|
\bibitem[Hu et~al.(2018)]{hu2018squeeze}
|
|
Hu, J., Shen, L., and Sun, G.
|
|
\newblock Squeeze-and-excitation networks.
|
|
\newblock \emph{CVPR}, pp. 7132--7141, 2018.
|
|
|
|
\bibitem[Hu et~al.(2022)]{hu2022pushing}
|
|
Hu, S.X., Li, D., St{\"u}hmer, J., Kim, M., and Hospedales, T.M.
|
|
\newblock Pushing the limits of simple pipelines for few-shot learning.
|
|
\newblock \emph{CVPR}, pp. 9068--9077, 2022.
|
|
|
|
\bibitem[Kahl et~al.(2021)]{kahl2021birdnet}
|
|
Kahl, S., Wood, C.M., Eibl, M., and Klinck, H.
|
|
\newblock BirdNET: A deep learning solution for avian diversity monitoring.
|
|
\newblock \emph{Ecological Informatics}, 61:101236, 2021.
|
|
|
|
\bibitem[Li et~al.(2019)]{li2019large}
|
|
Li, A., Luo, T., Lu, Z., Xiang, T., and Wang, L.
|
|
\newblock Large-scale few-shot learning: Knowledge transfer with class hierarchy.
|
|
\newblock \emph{CVPR}, pp. 7212--7220, 2019.
|
|
|
|
\bibitem[Lin et~al.(2015)]{lin2015bilinear}
|
|
Lin, T.Y., RoyChowdhury, A., and Maji, S.
|
|
\newblock Bilinear CNN models for fine-grained visual recognition.
|
|
\newblock \emph{ICCV}, pp. 1449--1457, 2015.
|
|
|
|
\bibitem[Liu et~al.(2020)]{liu2020many}
|
|
Liu, B., Kang, H., Li, H., Hua, G., and Vasconcelos, N.
|
|
\newblock Few-shot open-set recognition using meta-learning.
|
|
\newblock \emph{CVPR}, pp. 8798--8807, 2020.
|
|
|
|
\bibitem[Lorieul et~al.(2022)]{lorieul2022overview}
|
|
Lorieul, T., Cole, E., Deneu, B., et~al.
|
|
\newblock Overview of GeoLifeCLEF 2022: Predicting species presence.
|
|
\newblock \emph{CLEF}, 2022.
|
|
|
|
\bibitem[Norouzzadeh et~al.(2018)]{norouzzadeh2018automatically}
|
|
Norouzzadeh, M.S., Nguyen, A., Kosmala, M., et~al.
|
|
\newblock Automatically identifying wild animals in camera-trap images.
|
|
\newblock \emph{PNAS}, 115(25):E5716--E5725, 2018.
|
|
|
|
\bibitem[Oquab et~al.(2024)]{oquab2024dinov2}
|
|
Oquab, M., Darcet, T., Moutakanni, T., et~al.
|
|
\newblock DINOv2: Learning robust visual features without supervision.
|
|
\newblock \emph{TMLR}, 2024.
|
|
|
|
\bibitem[Snell et~al.(2017)]{snell2017prototypical}
|
|
Snell, J., Swersky, K., and Zemel, R.
|
|
\newblock Prototypical networks for few-shot learning.
|
|
\newblock \emph{NeurIPS}, pp. 4077--4087, 2017.
|
|
|
|
\bibitem[Steenweg et~al.(2017)]{steenweg2017scaling}
|
|
Steenweg, R., Hebblewhite, M., Kays, R., et~al.
|
|
\newblock Scaling-up camera traps: monitoring biodiversity with sensor networks.
|
|
\newblock \emph{Frontiers in Ecology and the Environment}, 15(1):26--34, 2017.
|
|
|
|
\bibitem[Sung et~al.(2018)]{sung2018learning}
|
|
Sung, F., Yang, Y., Zhang, L., et~al.
|
|
\newblock Learning to compare: Relation network for few-shot learning.
|
|
\newblock \emph{CVPR}, pp. 1199--1208, 2018.
|
|
|
|
\bibitem[Van~Horn et~al.(2015)]{van2015building}
|
|
Van~Horn, G., Branson, S., Farrell, R., et~al.
|
|
\newblock Building a bird recognition app with citizen scientists.
|
|
\newblock \emph{CVPR}, pp. 595--604, 2015.
|
|
|
|
\bibitem[Van~Horn et~al.(2021)]{van2021benchmarking}
|
|
Van~Horn, G., Cole, E., Beery, S., et~al.
|
|
\newblock Benchmarking representation learning for natural world images.
|
|
\newblock \emph{CVPR}, pp. 12884--12893, 2021.
|
|
|
|
\bibitem[Vinyals et~al.(2016)]{vinyals2016matching}
|
|
Vinyals, O., Blundell, C., Lillicrap, T., and Welling, M.
|
|
\newblock Matching networks for one shot learning.
|
|
\newblock \emph{NeurIPS}, pp. 3630--3638, 2016.
|
|
|
|
\bibitem[Wang et~al.(2020)]{wang2020generalizing}
|
|
Wang, Y., Yao, Q., Kwok, J.T., and Ni, L.M.
|
|
\newblock Generalizing from a few examples: A survey on few-shot learning.
|
|
\newblock \emph{ACM Computing Surveys}, 53(3):1--34, 2020.
|
|
|
|
\bibitem[Wei et~al.(2019)]{wei2019piecewise}
|
|
Wei, X.S., Wang, P., Liu, L., Shen, C., and Wu, J.
|
|
\newblock Piecewise classifier mappings for few-shot fine-grained recognition.
|
|
\newblock \emph{IEEE TIP}, 28(12):6116--6125, 2019.
|
|
|
|
\bibitem[Ye et~al.(2020)]{ye2020few}
|
|
Ye, H.J., Hu, H., Zhan, D.C., and Sha, F.
|
|
\newblock Few-shot learning via saliency-guided hallucination of samples.
|
|
\newblock \emph{CVPR}, pp. 770--779, 2020.
|
|
|
|
\bibitem[Zhang et~al.(2020)]{zhang2020deepemd}
|
|
Zhang, C., Cai, Y., Lin, G., and Shen, C.
|
|
\newblock DeepEMD: Few-shot image classification with differentiable EMD.
|
|
\newblock \emph{CVPR}, pp. 12203--12213, 2020.
|
|
|
|
\bibitem[Zheng et~al.(2017)]{zheng2017learning}
|
|
Zheng, H., Fu, J., Mei, T., and Luo, J.
|
|
\newblock Learning multi-attention CNN for fine-grained image recognition.
|
|
\newblock \emph{ICCV}, pp. 5209--5217, 2017.
|
|
|
|
\end{thebibliography}
|
|
|
|
% ══════════════════════════════════════════════════════════════════════════════
|
|
\appendix
|