{ "id": 82437, "title": "Knowledge roundup - papers and resources shared by people.", "url": "/competitions/humpback-whale-identification/discussion/82437", "author_name": "", "post_date": "2019-03-01T10:51:45.429402600Z", "votes": 23, "comment_count": 7, "views": 0, "content": "

I first want to thank Kaggle and Happywhale for organizing this interesting competition. Congratulations to the winners and all participants.

\n\n

Between Quora Insincere Questions Classification and Elo Merchant Category Recommendation competitions, I really did not have much time but joined this competition in the last 12 days to try out a couple of simple solutions and learn a few things. In the end my result is the average result of @daisukelab's ProtoNet (Thank you) and a simple siamese network. But I got lots of reading material in terms of papers and tools posted, I am so glad I joined. Thanks everyone who posted and shared in the amazing discussions.

\n\n

So I thought it would be a good idea to put all the papers and tools shared in one place for future visitors. In addition to this I suggest for people to read the top solutions posted and all of Heng's posts as I find them very informative.

\n\n\n\n

Please if you think I missed something let me know below and I can add it to the list.

", "messages": [ { "id": "481397", "postDate": "03/01/2019 10:51:45", "content": "

I first want to thank Kaggle and Happywhale for organizing this interesting competition. Congratulations to the winners and all participants.

\n\n

Between Quora Insincere Questions Classification and Elo Merchant Category Recommendation competitions, I really did not have much time but joined this competition in the last 12 days to try out a couple of simple solutions and learn a few things. In the end my result is the average result of @daisukelab's ProtoNet (Thank you) and a simple siamese network. But I got lots of reading material in terms of papers and tools posted, I am so glad I joined. Thanks everyone who posted and shared in the amazing discussions.

\n\n

So I thought it would be a good idea to put all the papers and tools shared in one place for future visitors. In addition to this I suggest for people to read the top solutions posted and all of Heng's posts as I find them very informative.

\n\n\n\n

Please if you think I missed something let me know below and I can add it to the list.

", "rawMarkdown": "I first want to thank Kaggle and Happywhale for organizing this interesting competition. Congratulations to the winners and all participants.\n\nBetween [Quora Insincere Questions Classification][1] and [Elo Merchant Category Recommendation][2] competitions, I really did not have much time but joined this competition in the last 12 days to try out a couple of simple solutions and learn a few things. In the end my result is the average result of @daisukelab's [ProtoNet][3] (Thank you) and a simple siamese network. But I got lots of reading material in terms of papers and tools posted, I am so glad I joined. Thanks everyone who posted and shared in the amazing discussions.\n\nSo I thought it would be a good idea to put all the papers and tools shared in one place for future visitors. In addition to this I suggest for people to read the top solutions posted and all of Heng's posts as I find them very informative.\n\n - **Thanks Earhian aka @qiaojian**, the [winner][4] said - For few shot learning, my method is from https://arxiv.org/abs/1707.05574. I used heavy augment and class-balanced sampler.\n\n - **Thanks Heng aka @hengck23** \n- https://arxiv.org/pdf/1803.00676.pdf \n- \"IMAGE DEFORMATION META-NETWORKS FOR ONESHOT LEARNING\"\nhttps://openreview.net/pdf?id=Sylw7nCqFQ \n- (another paper with almost same idea and same results: http://yugangjiang.info/publication/19AAAI-oneshot.pdf) \n\"Semi-Supervised Few-Shot Learning with Prototypical Networks\"\nhttp://metalearning.ml/2017/papers/metalearn17_boney.pdf \n- SphereFace: Deep Hypersphere Embedding for Face Recognition https://arxiv.org/abs/1704.08063 \n- \"Alpha pooling for fine-grained recognition\"\nhttps://github.com/cvjena/alpha_pooling \n- Large-scale Bisample Learning on ID Versus Spot Face\nRecognition - https://arxiv.org/pdf/1806.03018.pdf \n-\"Fix Your Features: Stationary and Maximally Discriminative Embeddings using\nRegular Polytope (Fixed Classifier) Networks\" - https://arxiv.org/pdf/1902.10441.pdf \n- \"Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet\"\nhttps://openreview.net/forum?id=SkfMWhAqYQ\n-\"Bags of Local Convolutional Features for Scalable Instance Search\" - https://core.ac.uk/download/pdf/147609064.pdf \n- The Unreasonable Effectiveness of Deep Features as a Perceptual Metric - https://github.com/richzhang/PerceptualSimilarity \n- Deep Deformation Network for Object Landmark Localization - NEC paper\nhttps://arxiv.org/pdf/1605.01014\n\n\n - **Thanks @pheadrus** - https://www.youtube.com/watch?v=wcKL05DomBU\n\n - **Thanks @jeandebleau** \n- Gaussian version of the prototypical network : https://arxiv.org/abs/1708.02735 \n- http://openaccess.thecvf.com/content_ICCV_2017/papers/Movshovitz-Attias_No_Fuss_Distance_ICCV_2017_paper.pdf\n- \"soft k nearest neighbor loss\", and especially compare with triplet loss: \nhttps://arxiv.org/pdf/1902.01889.pdf\n\n - **Thanks Eduardo aka @arc144** \n- This paper here https://arxiv.org/pdf/1803.00676.pdf uses new entries in training as a type of semi-supervised ProtoNet.\n- Fine-tuning CNN Image Retrieval with No Human Annotation - https://arxiv.org/pdf/1711.02512.pdf\n\n - **Thanks Miguel aka @mnpinto** - Re-ranking Person Re-identification with k-reciprocal Encoding - https://arxiv.org/pdf/1701.08398.pdf\n\n - **Thanks @daisukelab** \n- Advances in few-shot learning: a guided tour - https://towardsdatascience.com/advances-in-few-shot-learning-a-guided-tour-36bc10a68b77\n- Original paper: Prototpyical Networks for Few-shot Learning, Snell et al.\nhttps://arxiv.org/pdf/1703.05175.pdf\n\n - **Meta-Learning to Make Smart Inferences from Small Data** - https://www.youtube.com/watch?v=NpSpHlHpz6k\n\n - **On First-Order Meta-Learning Algorithms** - https://arxiv.org/pdf/1803.02999.pdf\n\nPlease if you think I missed something let me know below and I can add it to the list.\n\n\n [1]: https://www.kaggle.com/c/quora-insincere-questions-classification\n [2]: https://www.kaggle.com/c/elo-merchant-category-recommendation\n [3]: https://www.kaggle.com/c/humpback-whale-identification/discussion/81085\n [4]: https://www.kaggle.com/c/humpback-whale-identification/discussion/82366", "votes": null }, { "id": "481408", "postDate": "03/01/2019 11:18:28", "content": "

Thanks, I was planning to do the same thing too. It was happy competition that many people have shared idea. :)

", "rawMarkdown": "Thanks, I was planning to do the same thing too. It was happy competition that many people have shared idea. :)", "votes": null }, { "id": "481446", "postDate": "03/01/2019 12:13:23", "content": "

Thanks @daisukelab, very happy one indeed. There are so many useful materials and I had a chance to read only a few so far. Thanks again for sharing your repo.

", "rawMarkdown": "Thanks @daisukelab, very happy one indeed. There are so many useful materials and I had a chance to read only a few so far. Thanks again for sharing your repo.", "votes": null }, { "id": "481626", "postDate": "03/01/2019 16:18:44", "content": "

this is a valuable challenge because it did several things at one time, including classification, metric learning, few shot learning (KNN classifier), image retrieval and distractor/outlier rejection (aka one-class training).

\n\n

I highly recommend kagglers to repeat some of the top solutions, because these techniques are very useful for future competition.

\n\n

i myself will be making post-submission.

", "rawMarkdown": "this is a valuable challenge because it did several things at one time, including classification, metric learning, few shot learning (KNN classifier), image retrieval and distractor/outlier rejection (aka one-class training).\n\nI highly recommend kagglers to repeat some of the top solutions, because these techniques are very useful for future competition. \n\ni myself will be making post-submission.", "votes": null }, { "id": "481813", "postDate": "03/01/2019 22:08:10", "content": "

Thanks @hengck23 for all the education through all that you have shared here and in past competitions. I am with you on repeating the top solutions and the usefulness of the kowledge to be gained here. For me, the competition looked simple from a distance until I started reading through the discussions a couple of weeks ago and realised its uniqueness. Hence as late as I was to the game, I just had to jump in knowing I can learn so much here. So compiling this list is partly for myself as well to redo this after the competition to enhance my knowledge.

", "rawMarkdown": "Thanks @hengck23 for all the education through all that you have shared here and in past competitions. I am with you on repeating the top solutions and the usefulness of the kowledge to be gained here. For me, the competition looked simple from a distance until I started reading through the discussions a couple of weeks ago and realised its uniqueness. Hence as late as I was to the game, I just had to jump in knowing I can learn so much here. So compiling this list is partly for myself as well to redo this after the competition to enhance my knowledge.", "votes": null }, { "id": "482737", "postDate": "03/03/2019 15:27:05", "content": "

Thanks for the paper survey... You did a great job, and pretty sure it will be a good reference for future..\nThe link that I mentioned in one of my posts actually got it from the great protonet code shared in the discussion by @diasukelab . So the credit goes to him actually...

", "rawMarkdown": "Thanks for the paper survey... You did a great job, and pretty sure it will be a good reference for future..\nThe link that I mentioned in one of my posts actually got it from the great protonet code shared in the discussion by @diasukelab . So the credit goes to him actually...", "votes": null }, { "id": "482895", "postDate": "03/03/2019 20:00:09", "content": "

Thanks @hwasiti, I have amended my list by crediting @daisukelab's for that paper.

", "rawMarkdown": "Thanks @hwasiti, I have amended my list by crediting @daisukelab's for that paper.", "votes": null }, { "id": "482905", "postDate": "03/03/2019 20:23:43", "content": "

Hey @hwasiti and @sheriytm, I really don't care about that. Anyway I hope your great works in the future competitions!

", "rawMarkdown": "Hey @hwasiti and @sheriytm, I really don't care about that. Anyway I hope your great works in the future competitions!", "votes": null } ], "comments": [ { "id": 481408, "author_name": "daisukelab", "author_url": "", "post_date": "03/01/2019 11:18:28", "content": "

Thanks, I was planning to do the same thing too. It was happy competition that many people have shared idea. :)

", "votes": null, "replies": [ { "id": 481446, "author_name": "sheriytm", "author_url": "", "post_date": "03/01/2019 12:13:23", "content": "

Thanks @daisukelab, very happy one indeed. There are so many useful materials and I had a chance to read only a few so far. Thanks again for sharing your repo.

", "votes": null, "replies": [] } ] }, { "id": 481626, "author_name": "hengck23", "author_url": "", "post_date": "03/01/2019 16:18:44", "content": "

this is a valuable challenge because it did several things at one time, including classification, metric learning, few shot learning (KNN classifier), image retrieval and distractor/outlier rejection (aka one-class training).

\n\n

I highly recommend kagglers to repeat some of the top solutions, because these techniques are very useful for future competition.

\n\n

i myself will be making post-submission.

", "votes": null, "replies": [ { "id": 481813, "author_name": "sheriytm", "author_url": "", "post_date": "03/01/2019 22:08:10", "content": "

Thanks @hengck23 for all the education through all that you have shared here and in past competitions. I am with you on repeating the top solutions and the usefulness of the kowledge to be gained here. For me, the competition looked simple from a distance until I started reading through the discussions a couple of weeks ago and realised its uniqueness. Hence as late as I was to the game, I just had to jump in knowing I can learn so much here. So compiling this list is partly for myself as well to redo this after the competition to enhance my knowledge.

", "votes": null, "replies": [] } ] }, { "id": 482737, "author_name": "hwasiti", "author_url": "", "post_date": "03/03/2019 15:27:05", "content": "

Thanks for the paper survey... You did a great job, and pretty sure it will be a good reference for future..\nThe link that I mentioned in one of my posts actually got it from the great protonet code shared in the discussion by @diasukelab . So the credit goes to him actually...

", "votes": null, "replies": [ { "id": 482895, "author_name": "sheriytm", "author_url": "", "post_date": "03/03/2019 20:00:09", "content": "

Thanks @hwasiti, I have amended my list by crediting @daisukelab's for that paper.

", "votes": null, "replies": [] }, { "id": 482905, "author_name": "daisukelab", "author_url": "", "post_date": "03/03/2019 20:23:43", "content": "

Hey @hwasiti and @sheriytm, I really don't care about that. Anyway I hope your great works in the future competitions!

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "481397": "I first want to thank Kaggle and Happywhale for organizing this interesting competition. Congratulations to the winners and all participants.\n\nBetween [Quora Insincere Questions Classification][1] and [Elo Merchant Category Recommendation][2] competitions, I really did not have much time but joined this competition in the last 12 days to try out a couple of simple solutions and learn a few things. In the end my result is the average result of @daisukelab's [ProtoNet][3] (Thank you) and a simple siamese network. But I got lots of reading material in terms of papers and tools posted, I am so glad I joined. Thanks everyone who posted and shared in the amazing discussions.\n\nSo I thought it would be a good idea to put all the papers and tools shared in one place for future visitors. In addition to this I suggest for people to read the top solutions posted and all of Heng's posts as I find them very informative.\n\n - **Thanks Earhian aka @qiaojian**, the [winner][4] said - For few shot learning, my method is from https://arxiv.org/abs/1707.05574. I used heavy augment and class-balanced sampler.\n\n - **Thanks Heng aka @hengck23** \n- https://arxiv.org/pdf/1803.00676.pdf \n- \"IMAGE DEFORMATION META-NETWORKS FOR ONESHOT LEARNING\"\nhttps://openreview.net/pdf?id=Sylw7nCqFQ \n- (another paper with almost same idea and same results: http://yugangjiang.info/publication/19AAAI-oneshot.pdf) \n\"Semi-Supervised Few-Shot Learning with Prototypical Networks\"\nhttp://metalearning.ml/2017/papers/metalearn17_boney.pdf \n- SphereFace: Deep Hypersphere Embedding for Face Recognition https://arxiv.org/abs/1704.08063 \n- \"Alpha pooling for fine-grained recognition\"\nhttps://github.com/cvjena/alpha_pooling \n- Large-scale Bisample Learning on ID Versus Spot Face\nRecognition - https://arxiv.org/pdf/1806.03018.pdf \n-\"Fix Your Features: Stationary and Maximally Discriminative Embeddings using\nRegular Polytope (Fixed Classifier) Networks\" - https://arxiv.org/pdf/1902.10441.pdf \n- \"Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet\"\nhttps://openreview.net/forum?id=SkfMWhAqYQ\n-\"Bags of Local Convolutional Features for Scalable Instance Search\" - https://core.ac.uk/download/pdf/147609064.pdf \n- The Unreasonable Effectiveness of Deep Features as a Perceptual Metric - https://github.com/richzhang/PerceptualSimilarity \n- Deep Deformation Network for Object Landmark Localization - NEC paper\nhttps://arxiv.org/pdf/1605.01014\n\n\n - **Thanks @pheadrus** - https://www.youtube.com/watch?v=wcKL05DomBU\n\n - **Thanks @jeandebleau** \n- Gaussian version of the prototypical network : https://arxiv.org/abs/1708.02735 \n- http://openaccess.thecvf.com/content_ICCV_2017/papers/Movshovitz-Attias_No_Fuss_Distance_ICCV_2017_paper.pdf\n- \"soft k nearest neighbor loss\", and especially compare with triplet loss: \nhttps://arxiv.org/pdf/1902.01889.pdf\n\n - **Thanks Eduardo aka @arc144** \n- This paper here https://arxiv.org/pdf/1803.00676.pdf uses new entries in training as a type of semi-supervised ProtoNet.\n- Fine-tuning CNN Image Retrieval with No Human Annotation - https://arxiv.org/pdf/1711.02512.pdf\n\n - **Thanks Miguel aka @mnpinto** - Re-ranking Person Re-identification with k-reciprocal Encoding - https://arxiv.org/pdf/1701.08398.pdf\n\n - **Thanks @daisukelab** \n- Advances in few-shot learning: a guided tour - https://towardsdatascience.com/advances-in-few-shot-learning-a-guided-tour-36bc10a68b77\n- Original paper: Prototpyical Networks for Few-shot Learning, Snell et al.\nhttps://arxiv.org/pdf/1703.05175.pdf\n\n - **Meta-Learning to Make Smart Inferences from Small Data** - https://www.youtube.com/watch?v=NpSpHlHpz6k\n\n - **On First-Order Meta-Learning Algorithms** - https://arxiv.org/pdf/1803.02999.pdf\n\nPlease if you think I missed something let me know below and I can add it to the list.\n\n\n [1]: https://www.kaggle.com/c/quora-insincere-questions-classification\n [2]: https://www.kaggle.com/c/elo-merchant-category-recommendation\n [3]: https://www.kaggle.com/c/humpback-whale-identification/discussion/81085\n [4]: https://www.kaggle.com/c/humpback-whale-identification/discussion/82366", "481408": "Thanks, I was planning to do the same thing too. It was happy competition that many people have shared idea. :)", "481446": "Thanks @daisukelab, very happy one indeed. There are so many useful materials and I had a chance to read only a few so far. Thanks again for sharing your repo.", "481626": "this is a valuable challenge because it did several things at one time, including classification, metric learning, few shot learning (KNN classifier), image retrieval and distractor/outlier rejection (aka one-class training).\n\nI highly recommend kagglers to repeat some of the top solutions, because these techniques are very useful for future competition. \n\ni myself will be making post-submission.", "481813": "Thanks @hengck23 for all the education through all that you have shared here and in past competitions. I am with you on repeating the top solutions and the usefulness of the kowledge to be gained here. For me, the competition looked simple from a distance until I started reading through the discussions a couple of weeks ago and realised its uniqueness. Hence as late as I was to the game, I just had to jump in knowing I can learn so much here. So compiling this list is partly for myself as well to redo this after the competition to enhance my knowledge.", "482737": "Thanks for the paper survey... You did a great job, and pretty sure it will be a good reference for future..\nThe link that I mentioned in one of my posts actually got it from the great protonet code shared in the discussion by @diasukelab . So the credit goes to him actually...", "482895": "Thanks @hwasiti, I have amended my list by crediting @daisukelab's for that paper.", "482905": "Hey @hwasiti and @sheriytm, I really don't care about that. Anyway I hope your great works in the future competitions!" }, "source": "meta" }