A Reproducible Resource-Aware Cascade for Wildlife Image Analysis Under Limited Data

Xixian Wang

doi:10.6911/

Authors

Xixian Wang The Ohio State University, Ohio, USA

DOI:

https://doi.org/10.6911/

Keywords:

Camera traps; wildlife image classification; conditional inference; model cascade; confidence calibration; group-stratified cross-validation; resource-aware inference.

Abstract

Selecting when to invoke a costly classifier is harder than evaluating the cost-performance curve after the fact. We examine this problem in a reproducible wildlife image-analysis prototype that uses a lightweight model for routine screening and a stronger model for selected cases. The evaluation uses 1,000 verified ENA24 images from four balanced classes. Numeric-ID burst proxies and identical 64-bit difference hashes produce 160 conservative groups. Five outer group-stratified folds keep fitting, model selection, calibration, routing validation, and testing separate. Frozen ImageNet-pretrained MobileNetV3-Small and ResNet-18 extract features for multinomial logistic-regression heads. Temperature scaling is fitted on calibration data. Each fold then selects the lowest-cost threshold whose routing-validation Macro-F1 is within 0.02 of stage 2. The pooled out-of-fold Macro-F1 is 0.89535 for stage 1 and 0.92397 for stage 2. The validation-selected cascade calls stage 2 for 3.9% of test images and reaches 0.89720, a 0.02677 gap from stage 2 that exceeds the specified tolerance. Higher fixed budgets approach stage-2 performance, but they do not change the failure of the prospective selection rule. The result points to threshold-transfer instability in this small-data setting. Acoustic or other contextual signals are a possible extension, not an experiment reported here. The resulting artifact is a reproducible visual prototype with an explicit evaluation protocol and a clearly reported negative result.

Downloads

Download data is not yet available.

References

[1] Swanson, A., Kosmala, M., Lintott, C., Simpson, R., Smith, A., & Packer, C. (2015). Snapshot Serengeti, high-frequency annotated camera trap images of 40 mammalian species in an African savanna. Scientific Data, 2, 150026. https://doi.org/10.1038/sdata.2015.26.

[2] Labeled Information Library of Alexandria: Biology and Conservation. (2019). ENA24-detection. LILA BC dataset page. https://lila.science/datasets/ena24detection

[3] Norouzzadeh, M. S., et al. (2018). Automatically identifying, counting, and describing wild animals in camera-trap images with deep learning. Proceedings of the National Academy of Sciences, 115(25), E5716–E5725. https://doi.org/10.1073/pnas.1719367115

[4] Yousif, H., Yuan, J., Kays, R., & He, Z. (2017). Fast human-animal detection from highly cluttered camera-trap images using joint background modeling and deep learning classification. In 2017 IEEE International Symposium on Circuits and Systems (ISCAS) (pp. 1–4). https://doi.org/10.1109/ISCAS.2017.8050762

[5] Yousif, H., Yuan, J., Kays, R., & He, Z. (2019). Animal scanner: Software for classifying humans, animals, and empty frames in camera trap images. Ecology and Evolution, 9(4), 1578–1589. https://doi.org/10.1002/ece3.4747

[6] Teerapittayanon, S., McDanel, B., & Kung, H. T. (2016). BranchyNet: Fast inference via early exiting from deep neural networks. In 2016 23rd International Conference on Pattern Recognition (ICPR) (pp. 2464–2469). https://doi.org/10.1109/ICPR.2016.7900006

[7] Bolukbasi, T., Wang, J., Dekel, O., & Saligrama, V. (2017). Adaptive neural networks for efficient inference. In Proceedings of the 34th International Conference on Machine Learning (Vol. 70, pp. 527–536). PMLR.

[8] Wang, X., Yu, F., Dou, Z.-Y., Darrell, T., & Gonzalez, J. E. (2018). SkipNet: Learning dynamic routing in convolutional networks. In Proceedings of the European Conference on Computer Vision (ECCV) (pp. 420–436). https://doi.org/10.1007/978-3-030-01261-8_25

[9] Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. (2017). On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning (Vol. 70, pp. 1321–1330). PMLR.

[10] Nixon, J., Dusenberry, M. W., Zhang, L., Jerfel, G., & Tran, D. (2019). Measuring calibration in deep learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (pp. 38–41).

[11] Beery, S., Van Horn, G., & Perona, P. (2018). Recognition in terra incognita. In Proceedings of the European Conference on Computer Vision (ECCV) (pp. 472–489). https://doi.org/10.1007/978-3-030-01270-0_28

[12] Tabak, M. A., et al. (2020). Improving the accessibility and transferability of machine learning algorithms for identification of animals in camera trap images: MLWIC2. Ecology and Evolution, 10(19), 10374–10383. https://doi.org/10.1002/ece3.6692

[13] Howard, A., et al. (2019). Searching for MobileNetV3. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) (pp. 1314–1324). https://doi.org/10.1109/ICCV.2019.00140

[14] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 770–778). https://doi.org/10.1109/CVPR.2016.90

[15] Kline, J., Potlapally, A., Pillai, B., Wani, T., Patil, V., Covey, P., Katole, R., Stevens, S., Subramoni, H., Berger-Wolf, T., & Stewart, C. (2025). SmartWilds: Multimodal wildlife monitoring dataset. arXiv preprint arXiv:2509.18894 (Version 2). https://arxiv.org/abs/2509.18894v2

[16] Pillai, B., Viswapriyan, V., Stewart, C., Berger-Wolf, T., & Kline, J. (2026). Cross-modal corroboration for annotation-free wildlife monitoring. arXiv preprint arXiv:2606.21613 (Version 1). https://arxiv.org/abs/2606.21613v1

[17] Yousif, H., Kays, R., & He, Z. (2019). Dynamic programming selection of object proposals for sequence-level animal species classification in the wild. IEEE Transactions on Circuits and Systems for Video Technology.