Hi everyone,
\nWe are happy to announce our fourth annual herbarium fine-grained recognition challenge, in partnership with the New York Botanical Garden, integral to a project to automate species recognition and aid in species discovery. The challenge is part of the Ninth Workshop on Fine-Grained Visual Categorization (FGVC9) at CVPR 2022.
\nOur first challenge, Herbarium 2019, focused on a dataset of 700 taxa from the well-curated flowering plant family, Melastomataceae, with all images coming from the NYBG virtual herbarium. Our second challenge, Herbarium 2020, expanded the dataset outside of a single family, spanning 32K vascular plant taxa from the virtual herbarium. Our third challenge, Herbarium 2021, included over 65K vascular plants with 2.5 M images gathered from five different institutions.
\nFor the herbarium 2022 competition, we attempted to build a dataset that can be the basis of a useful identification tool: it covers all recognized vascular plant forms so far documented in the scientific literature for a particular geographical region. Our dataset encompasses over 90% of all plant taxa in North America, spanning 1.05M images for 15501 plant taxa, collected from 60 different institutions around the world. In addition to the images, we also included hierarchical taxonomic structure, and for the first time, a set of pairwise phylogenetic distances among taxa this year, in the hopes that knowledge of plant systematics may provide some prior knowledge useful for guiding visual representation learning towards the right direction.
\nFor those who are interested in looking at the previous competitions more closely for learning purposes, here is our most recent publication on the herbarium challenge. From the paper, I think the most useful thing to share may be this table that summarizes the model architectures and settings that winning teams employed at herbarium 2021.
\nWe understand from previous herbarium competitions that large numbers of images may be an obstacle for participation. To encourage more teams to participate this year, we set the maximum number of training images to 80 images per taxon, compared to the last year's competition in which the maximum was up to several thousand per taxon. Hopefully the training process will not be too painful and all teams could have some fun with this unique dataset!
\nLooking forward to the interesting and creative solutions that you all come up with. Feel free to ask questions here, or on the github page, which will be updated soon for this year's competition.
\nWe are currently working on the baseline model. We plan on posting the notebook for the baseline soon. Stay tuned!
", "messages": [ { "id": 1691848, "postDate": "2022-02-15T16:41:39.847Z", "content": "Hi everyone,
\nWe are happy to announce our fourth annual herbarium fine-grained recognition challenge, in partnership with the New York Botanical Garden, integral to a project to automate species recognition and aid in species discovery. The challenge is part of the Ninth Workshop on Fine-Grained Visual Categorization (FGVC9) at CVPR 2022.
\nOur first challenge, Herbarium 2019, focused on a dataset of 700 taxa from the well-curated flowering plant family, Melastomataceae, with all images coming from the NYBG virtual herbarium. Our second challenge, Herbarium 2020, expanded the dataset outside of a single family, spanning 32K vascular plant taxa from the virtual herbarium. Our third challenge, Herbarium 2021, included over 65K vascular plants with 2.5 M images gathered from five different institutions.
\nFor the herbarium 2022 competition, we attempted to build a dataset that can be the basis of a useful identification tool: it covers all recognized vascular plant forms so far documented in the scientific literature for a particular geographical region. Our dataset encompasses over 90% of all plant taxa in North America, spanning 1.05M images for 15501 plant taxa, collected from 60 different institutions around the world. In addition to the images, we also included hierarchical taxonomic structure, and for the first time, a set of pairwise phylogenetic distances among taxa this year, in the hopes that knowledge of plant systematics may provide some prior knowledge useful for guiding visual representation learning towards the right direction.
\nFor those who are interested in looking at the previous competitions more closely for learning purposes, here is our most recent publication on the herbarium challenge. From the paper, I think the most useful thing to share may be this table that summarizes the model architectures and settings that winning teams employed at herbarium 2021.
\nWe understand from previous herbarium competitions that large numbers of images may be an obstacle for participation. To encourage more teams to participate this year, we set the maximum number of training images to 80 images per taxon, compared to the last year's competition in which the maximum was up to several thousand per taxon. Hopefully the training process will not be too painful and all teams could have some fun with this unique dataset!
\nLooking forward to the interesting and creative solutions that you all come up with. Feel free to ask questions here, or on the github page, which will be updated soon for this year's competition.
\nWe are currently working on the baseline model. We plan on posting the notebook for the baseline soon. Stay tuned!
", "rawMarkdown": "Hi everyone,\n\nWe are happy to announce our fourth annual herbarium fine-grained recognition challenge, in partnership with the [New York Botanical Garden](https://www.nybg.org/), integral to a project to automate species recognition and aid in species discovery. The challenge is part of the [Ninth Workshop on Fine-Grained Visual Categorization (FGVC9)](https://sites.google.com/view/fgvc9/home) at [CVPR 2022](https://cvpr2022.thecvf.com/).\n\nOur first challenge, [Herbarium 2019](https://www.kaggle.com/c/herbarium-2019-fgvc6), focused on a dataset of 700 taxa from the well-curated flowering plant family, *Melastomataceae*, with all images coming from the [NYBG virtual herbarium](https://www.nybg.org/). Our second challenge, [Herbarium 2020](https://www.kaggle.com/c/herbarium-2020-fgvc7/overview), expanded the dataset outside of a single family, spanning 32K vascular plant taxa from the virtual herbarium. Our third challenge, [Herbarium 2021](https://www.kaggle.com/c/herbarium-2021-fgvc8), included over 65K vascular plants with 2.5 M images gathered from five different institutions. \n\nFor the herbarium 2022 competition, we attempted to build a dataset that can be the basis of a useful identification tool: it covers all recognized vascular plant forms so far documented in the scientific literature for a particular geographical region. Our dataset encompasses over 90% of all plant taxa in North America, spanning 1.05M images for 15501 plant taxa, collected from 60 different institutions around the world. In addition to the images, we also included hierarchical taxonomic structure, and for the first time, a set of pairwise phylogenetic distances among taxa this year, in the hopes that knowledge of plant systematics may provide some prior knowledge useful for guiding visual representation learning towards the right direction.\n\nFor those who are interested in looking at the previous competitions more closely for learning purposes, [here](https://www.frontiersin.org/articles/10.3389/fpls.2021.787127/full?&utm_source=Email_to_authors_&utm_medium=Email&utm_content=T1_11.5e1_author&utm_campaign=Email_publication&field=&journalName=Frontiers_in_Plant_Science&id=787127) is our most recent publication on the herbarium challenge. From the paper, I think the most useful thing to share may be this [table](https://www.frontiersin.org/files/Articles/787127/fpls-12-787127-HTML-r1/image_m/fpls-12-787127-t002.jpg) that summarizes the model architectures and settings that winning teams employed at herbarium 2021. \n\nWe understand from previous herbarium competitions that large numbers of images may be an obstacle for participation. To encourage more teams to participate this year, we set the maximum number of training images to 80 images per taxon, compared to the last year's competition in which the maximum was up to several thousand per taxon. Hopefully the training process will not be too painful and all teams could have some fun with this unique dataset!\n\nLooking forward to the interesting and creative solutions that you all come up with. Feel free to ask questions here, or on the [github page](https://github.com/visipedia/herbarium_comp), which will be updated soon for this year's competition.\n\nWe are currently working on the baseline model. We plan on posting the notebook for the baseline soon. Stay tuned!", "votes": 13 }, { "id": 1693305, "postDate": "2022-02-16T15:26:08.117Z", "content": "As mentioned in CVPR 2022 tab,
\na subset may be invited to present their results at the workshop
\nWill the teams be invited for poster presentation, or will they get a chance to have a workshop paper published?
Hi @Prateek, things are still up in the air; we are not sure yet. However, let me keep things posted when it gets more clear how we are going to pursue this year!
", "rawMarkdown": "Hi @Prateek, things are still up in the air; we are not sure yet. However, let me keep things posted when it gets more clear how we are going to pursue this year! ", "votes": 3 }, { "id": 1737289, "postDate": "2022-03-28T09:59:15.767Z", "content": "Hi John,
\nAny confirmation about the above question? Will the teams be invited for poster presentation, or will they get a chance to have a workshop paper published?
Hi @ks2019, still haven't hear anything about it yet. Let me keep things posted!
", "rawMarkdown": "Hi @ks2019, still haven't hear anything about it yet. Let me keep things posted!", "votes": 1 } ] }, { "id": 1692506, "postDate": "2022-02-16T04:39:49.483Z", "content": "Thanks for organising this challenge.😁😁😁👍👍👍
", "rawMarkdown": "Thanks for organising this challenge.😁😁😁👍👍👍", "votes": 1, "replies": [ { "id": 1693555, "postDate": "2022-02-16T18:25:19.700Z", "content": "Thank you for participating in this challenge!
", "rawMarkdown": "Thank you for participating in this challenge! ", "votes": 1 } ] }, { "id": 1691881, "postDate": "2022-02-15T17:02:30.030Z", "content": "Interesting competition. I think the idea of setting a maximum amount of images per Taxa makes sense and I hope it achieves your aims. I am excited to look in on this one.
", "rawMarkdown": "Interesting competition. I think the idea of setting a maximum amount of images per Taxa makes sense and I hope it achieves your aims. I am excited to look in on this one. ", "votes": 1, "replies": [ { "id": 1692180, "postDate": "2022-02-15T22:03:49.123Z", "content": "Thank you, Looking forward to see the outcome!
", "rawMarkdown": "Thank you, Looking forward to see the outcome! ", "votes": 1 } ] }, { "id": 1691883, "postDate": "2022-02-15T17:03:08.087Z", "content": "Thank you so much for the wonderful comp every year!
\nI had a question about why host the competition on Kaggle, I'm obviously a fan of Kaggle but I see many conferences host sprints or competitions on their pages.
\nHave you found the Kaggle solutions to be more interesting, or maybe increased participation, etc?
\nTIA!
", "rawMarkdown": "Thank you so much for the wonderful comp every year! \n\nI had a question about why host the competition on Kaggle, I'm obviously a fan of Kaggle but I see many conferences host sprints or competitions on their pages. \n\nHave you found the Kaggle solutions to be more interesting, or maybe increased participation, etc?\n\nTIA!", "replies": [ { "id": 1692182, "postDate": "2022-02-15T22:05:45.683Z", "content": "Hi @init27, you are welcome! Hopefully this one is fun to run!
\nI am not sure of the reason why CVPR hosts competitions through Kaggle. It would be great if anyone who knows about the history better could chime in!
", "rawMarkdown": "Hi @init27, you are welcome! Hopefully this one is fun to run!\n\nI am not sure of the reason why CVPR hosts competitions through Kaggle. It would be great if anyone who knows about the history better could chime in!", "votes": 2 } ] }, { "id": 1748341, "postDate": "2022-04-07T13:44:24.510Z", "content": "Hey there,
\nthx for your magnificent works!
But still, I have one question:
\nAccording to your description, we have about 15k taxa to be classed. Whereas in your \"sample submission CSV\", the prediction is 42 which seems like there are some discrepancies with your description. So, could you plz tell me how many classes should I predict with my model?
Looking forward to your reply!
\nCheers,
\nEnnard
In case anyone is interested in the plant science part, here is a brief introduction of the history of plant discovery and identification by one of the directors of New York Botanical Garden: FGVC7: Barbara Ambrose - Wait. You Do What At A Botanical Garden?
", "rawMarkdown": "In case anyone is interested in the plant science part, here is a brief introduction of the history of plant discovery and identification by one of the directors of New York Botanical Garden: [FGVC7: Barbara Ambrose - Wait. You Do What At A Botanical Garden?](https://www.youtube.com/watch?v=50E39xOPV5U)" } ], "comments": [ { "id": 1693305, "author_name": "Prateek", "author_url": "", "post_date": "2022-02-16T15:26:08.117000", "content": "As mentioned in CVPR 2022 tab,
\na subset may be invited to present their results at the workshop
\nWill the teams be invited for poster presentation, or will they get a chance to have a workshop paper published?
Hi @Prateek, things are still up in the air; we are not sure yet. However, let me keep things posted when it gets more clear how we are going to pursue this year!
", "votes": 3, "replies": [] }, { "id": 1737289, "author_name": "Kumar Shubham", "author_url": "", "post_date": "2022-03-28T09:59:15.767000", "content": "Hi John,
\nAny confirmation about the above question? Will the teams be invited for poster presentation, or will they get a chance to have a workshop paper published?
Hi @ks2019, still haven't hear anything about it yet. Let me keep things posted!
", "votes": 1, "replies": [] } ] }, { "id": 1692506, "author_name": "哇哈哈哈", "author_url": "", "post_date": "2022-02-16T04:39:49.483000", "content": "Thanks for organising this challenge.😁😁😁👍👍👍
", "votes": 1, "replies": [ { "id": 1693555, "author_name": "John Park", "author_url": "", "post_date": "2022-02-16T18:25:19.700000", "content": "Thank you for participating in this challenge!
", "votes": 1, "replies": [] } ] }, { "id": 1691881, "author_name": "Taran Marley", "author_url": "", "post_date": "2022-02-15T17:02:30.030000", "content": "Interesting competition. I think the idea of setting a maximum amount of images per Taxa makes sense and I hope it achieves your aims. I am excited to look in on this one.
", "votes": 1, "replies": [ { "id": 1692180, "author_name": "John Park", "author_url": "", "post_date": "2022-02-15T22:03:49.123000", "content": "Thank you, Looking forward to see the outcome!
", "votes": 1, "replies": [] } ] }, { "id": 1691883, "author_name": "Sanyam Bhutani", "author_url": "", "post_date": "2022-02-15T17:03:08.087000", "content": "Thank you so much for the wonderful comp every year!
\nI had a question about why host the competition on Kaggle, I'm obviously a fan of Kaggle but I see many conferences host sprints or competitions on their pages.
\nHave you found the Kaggle solutions to be more interesting, or maybe increased participation, etc?
\nTIA!
", "votes": 0, "replies": [ { "id": 1692182, "author_name": "John Park", "author_url": "", "post_date": "2022-02-15T22:05:45.683000", "content": "Hi @init27, you are welcome! Hopefully this one is fun to run!
\nI am not sure of the reason why CVPR hosts competitions through Kaggle. It would be great if anyone who knows about the history better could chime in!
", "votes": 2, "replies": [] } ] }, { "id": 1748341, "author_name": "Ennard Zhao", "author_url": "", "post_date": "2022-04-07T13:44:24.510000", "content": "Hey there,
\nthx for your magnificent works!
But still, I have one question:
\nAccording to your description, we have about 15k taxa to be classed. Whereas in your \"sample submission CSV\", the prediction is 42 which seems like there are some discrepancies with your description. So, could you plz tell me how many classes should I predict with my model?
Looking forward to your reply!
\nCheers,
\nEnnard
In case anyone is interested in the plant science part, here is a brief introduction of the history of plant discovery and identification by one of the directors of New York Botanical Garden: FGVC7: Barbara Ambrose - Wait. You Do What At A Botanical Garden?
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1691848": "Hi everyone,\n\nWe are happy to announce our fourth annual herbarium fine-grained recognition challenge, in partnership with the [New York Botanical Garden](https://www.nybg.org/), integral to a project to automate species recognition and aid in species discovery. The challenge is part of the [Ninth Workshop on Fine-Grained Visual Categorization (FGVC9)](https://sites.google.com/view/fgvc9/home) at [CVPR 2022](https://cvpr2022.thecvf.com/).\n\nOur first challenge, [Herbarium 2019](https://www.kaggle.com/c/herbarium-2019-fgvc6), focused on a dataset of 700 taxa from the well-curated flowering plant family, *Melastomataceae*, with all images coming from the [NYBG virtual herbarium](https://www.nybg.org/). Our second challenge, [Herbarium 2020](https://www.kaggle.com/c/herbarium-2020-fgvc7/overview), expanded the dataset outside of a single family, spanning 32K vascular plant taxa from the virtual herbarium. Our third challenge, [Herbarium 2021](https://www.kaggle.com/c/herbarium-2021-fgvc8), included over 65K vascular plants with 2.5 M images gathered from five different institutions. \n\nFor the herbarium 2022 competition, we attempted to build a dataset that can be the basis of a useful identification tool: it covers all recognized vascular plant forms so far documented in the scientific literature for a particular geographical region. Our dataset encompasses over 90% of all plant taxa in North America, spanning 1.05M images for 15501 plant taxa, collected from 60 different institutions around the world. In addition to the images, we also included hierarchical taxonomic structure, and for the first time, a set of pairwise phylogenetic distances among taxa this year, in the hopes that knowledge of plant systematics may provide some prior knowledge useful for guiding visual representation learning towards the right direction.\n\nFor those who are interested in looking at the previous competitions more closely for learning purposes, [here](https://www.frontiersin.org/articles/10.3389/fpls.2021.787127/full?&utm_source=Email_to_authors_&utm_medium=Email&utm_content=T1_11.5e1_author&utm_campaign=Email_publication&field=&journalName=Frontiers_in_Plant_Science&id=787127) is our most recent publication on the herbarium challenge. From the paper, I think the most useful thing to share may be this [table](https://www.frontiersin.org/files/Articles/787127/fpls-12-787127-HTML-r1/image_m/fpls-12-787127-t002.jpg) that summarizes the model architectures and settings that winning teams employed at herbarium 2021. \n\nWe understand from previous herbarium competitions that large numbers of images may be an obstacle for participation. To encourage more teams to participate this year, we set the maximum number of training images to 80 images per taxon, compared to the last year's competition in which the maximum was up to several thousand per taxon. Hopefully the training process will not be too painful and all teams could have some fun with this unique dataset!\n\nLooking forward to the interesting and creative solutions that you all come up with. Feel free to ask questions here, or on the [github page](https://github.com/visipedia/herbarium_comp), which will be updated soon for this year's competition.\n\nWe are currently working on the baseline model. We plan on posting the notebook for the baseline soon. Stay tuned!", "1693305": "As mentioned in CVPR 2022 tab,\n`a subset may be invited to present their results at the workshop`\nWill the teams be invited for poster presentation, or will they get a chance to have a workshop paper published?", "1692506": "Thanks for organising this challenge.😁😁😁👍👍👍", "1691881": "Interesting competition. I think the idea of setting a maximum amount of images per Taxa makes sense and I hope it achieves your aims. I am excited to look in on this one. ", "1691883": "Thank you so much for the wonderful comp every year! \n\nI had a question about why host the competition on Kaggle, I'm obviously a fan of Kaggle but I see many conferences host sprints or competitions on their pages. \n\nHave you found the Kaggle solutions to be more interesting, or maybe increased participation, etc?\n\nTIA!", "1748341": "Hey there,\nthx for your magnificent works!\n\nBut still, I have one question:\nAccording to your description, we have about 15k taxa to be classed. Whereas in your \"sample submission CSV\", the prediction is 42 which seems like there are some discrepancies with your description. So, could you plz tell me how many classes should I predict with my model?\n\nLooking forward to your reply!\n\nCheers,\nEnnard", "1702395": "In case anyone is interested in the plant science part, here is a brief introduction of the history of plant discovery and identification by one of the directors of New York Botanical Garden: [FGVC7: Barbara Ambrose - Wait. You Do What At A Botanical Garden?](https://www.youtube.com/watch?v=50E39xOPV5U)" } }