This repository provides a suite of pre-trained models designed to analyze high-resolution roadside images captured from a vehicle-mounted camera, specifically to detect the presence or absence of plant species recognized as invasive in Denmark.
\nThese models are made available to facilitate the replication of results from the paper “Adapting a Global Plant Identification Model to Detect Invasive Alien Plant Species in High-Resolution Roadside Images” [1]. The paper addresses the challenge of processing large images that may contain multiple invasive species simultaneously, including:
\n• Cytisus scoparius (L.) Link
\n• Lupinus polyphyllus Lindl.
\n• Pastinaca sativa L.
\n• Reynoutria spp. (Reynoutria japonica Houtt., Reynoutria sachalinensis (F.Schmidt) Nakai)
\n• Rosa rugosa Thunb.
\n• Solidago spp. (Solidago canadensis L., Solidago gigantea Aiton)
\n\n
The models were partially generated using the following code and data resources:
\n• Code: GitHub Repository https://github.com/plantnet/roadside-invasive-plant-identification
\n• Dataset: Zenodo Link https://zenodo.org/record/14013930 (DOI 10.5281/zenodo.14013930, accessed in October 2024)
\nThe code allows full reproduction of the results presented in the paper.
\n\n
All shared models are based on a BeiT architecture [2], pre-trained on a large dataset of plant images (6.5 million images initially covering 43,683 species from Pl@ntNet’s internal database [3]). The model output was subsequently reduced to the six target invasive species listed above, resulting in the file model_beit_danish.pth.
\nThis model may be useful in other projects aiming to identify any of these six species in standard-sized images showing individual plants. Basic code to test the model is provided in the file basic_usage_pretrained_model.py, which requires the installation of the timm library (version 0.9.6, August 2023).
\nThen, the shared code in GitHub Repository can adapt the model_beit_danish model to handle larger image resolutions and potentially multi-label images, following various methods discussed in the paper.
\nHere is a brief summary of the different models:
\n\n
1. Model File: model_beit_danish.pth
\n• Image Size: 384 x 384
\n• Class Count: 7
\n• Experiment ID in Article: XP3
\n• Notes: BEiT model pre-trained on ImageNet, then fine-tuned on Pl@ntNet data limited to six invasive plant species.
\n\n
2. Model File: model_vamis_state_dict_xp1.pth
\n• Image Size: 1024 x 768
\n• Class Count: 7
\n• Experiment ID in Article: XP1
\n• Notes: BEiT model with large input size, pre-trained on ImageNet and fine-tuned on roadside images.
\n\n
3. Model File: model_vamis_state_dict_xp2.pth
\n• Image Size: 1024 x 768
\n• Class Count: 7
\n• Experiment ID in Article: XP2
\n• Notes: BEiT model with large input size, pre-trained on ImageNet, then fine-tuned on Pl@ntNet and roadside images.
\n\n
4. Model File: model_tiling_state_dict_xp4.pth
\n• Image Size: 384 x 384
\n• Class Count: 6
\n• Experiment ID in Article: XP4
\n• Notes: Classifier head on the model_beit_danish backbone.
\n\n
5. Model File: model_tiling_state_dict_xp5h_BCE_LN__pickup_strategy.pth
\n• Image Size: 384 x 384
\n• Class Count: 6
\n• Experiment ID in Article: XP5
\n• Notes: Classifier head on the model_beit_danish backbone fine-tuned on roadside images.
\n\n
6. Model File: model_beit_danish_head.pth
\n• Image Size: 384 x 384
\n• Class Count: 6
\n• Experiment ID in Article: XP5
\n• Notes: Initial classifier head before fine-tuning on the model_beit_danish.pth backbone.
\n\n
7. Model File: model_beit_danish_with_vamis.pth
\n• Image Size: 1024 x 768
\n• Class Count: 7
\n• Experiment ID in Article: XP2
\n• Notes: Initial classifier head before fine-tuning on the model_vamis_state_dict_xp2.pth backbone.
\n\n
References:
\n[1] V. Espitalier, J.-C. Lombardo, H. Goëau, C. Botella, T.T. Høye, M. Dyrmannd, P. Bonnet, A. Joly, “Adapting a Global Plant Identification Model to Detect Invasive Alien Plant Species in High-Resolution Roadside Images” (submitted to Ecological Informatics).
\n[2] Bao, H., Dong, L., Piao, S., & Wei, F. (2021). “BEiT: BERT Pre-training of Image Transformers.” arXiv preprint arXiv:2106.08254.
\n[3] A. Affouard, H. Goeau, P. Bonnet, J.-C. Lombardo, A. Joly, “Pl@ntNet App in the Era of Deep Learning.” ICLR: International Conference on Learning Representations.
\n\n
The work described here was funded by the European Commission via the GUARDEN and MAMBO projects, which have received funding from the European Union’s Horizon Europe research and innovation programme under grant agreements 101060693 and 101060639. The opinions expressed in this work are those of the authors and are not necessarily those of the GUARDEN or MAMBO partners or the European Commission.
\nThis project was provided with computer and storage resources by GENCI at IDRIS thanks to the grant 2023- A0151011389 on the supercomputer Jean Zay's the A100 partition.
", "publication_date": "2024-10-30", "publisher": "Zenodo", "resource_type": { "id": "model", "title": { "de": "Modell", "en": "Model" } }, "rights": [ { "description": { "en": "The Creative Commons Attribution license allows re-distribution and re-use of a licensed work on the condition that the creator is appropriately credited." }, "icon": "cc-by-icon", "id": "cc-by-4.0", "props": { "scheme": "spdx", "url": "https://creativecommons.org/licenses/by/4.0/legalcode" }, "title": { "en": "Creative Commons Attribution 4.0 International" } } ], "title": "Image classification models dedicated to the identification of 6 invasive plant species in Denmark in roadside views" }, "parent": { "access": { "owned_by": { "user": "90498" }, "settings": { "accept_conditions_text": null, "allow_guest_requests": false, "allow_user_requests": false, "secret_link_expiration": 0 } }, "communities": {}, "id": "13891415", "pids": { "doi": { "client": "datacite", "identifier": "10.5281/zenodo.13891415", "provider": "datacite" } } }, "pids": { "doi": { "client": "datacite", "identifier": "10.5281/zenodo.13891416", "provider": "datacite" }, "oai": { "identifier": "oai:zenodo.org:13891416", "provider": "oai" } }, "revision_id": 4, "stats": { "all_versions": { "data_volume": 142375919152.0, "downloads": 835, "unique_downloads": 800, "unique_views": 370, "views": 385 }, "this_version": { "data_volume": 142375919152.0, "downloads": 835, "unique_downloads": 800, "unique_views": 370, "views": 385 } }, "status": "published", "swh": {}, "updated": "2024-10-30T18:30:26.396572+00:00", "versions": { "index": 1, "is_latest": true } }