hey guys, I'm using https://github.com/ageitgey/face_recognition for facial recognition. It seems to perform well on most videos from the training set except on the below videos. Anybody else facing similar issues with other face recognition libraries?
\n\nadhsbajydo\nebeknhudxq\nczmqpxrqoh\ncwqlvzefpg\nbbvgxeczei\navvdgsennp\ncycacemkmt\neoewqcpbgt\ndjvutyvaio\naxwgcsyphv\nesyhwdfnxs\ndhoqofwoxa
", "messages": [ { "id": 744848, "postDate": "2020-02-13T07:42:19.480Z", "content": "The ones I tried:
\n\n1) Uses MTCNN (Keras)
\n\nhttps://www.kaggle.com/unkownhihi/starter-kernel-with-cnn-model-ll-lb-0-69235
\n\n2) Uses MTCNN (torch)
\n\nhttps://www.kaggle.com/timesler/facial-recognition-model-in-pytorch
\n\n3) Uses Blazeface (torch)
\n\nhttps://www.kaggle.com/humananalog/inference-demo
", "rawMarkdown": "The ones I tried:\n\n1) Uses MTCNN (Keras)\n\nhttps://www.kaggle.com/unkownhihi/starter-kernel-with-cnn-model-ll-lb-0-69235\n\n2) Uses MTCNN (torch)\n\nhttps://www.kaggle.com/timesler/facial-recognition-model-in-pytorch\n\n3) Uses Blazeface (torch)\n\nhttps://www.kaggle.com/humananalog/inference-demo", "votes": 4 }, { "id": 744424, "postDate": "2020-02-12T20:31:22.657Z", "content": "hey guys, I'm using https://github.com/ageitgey/face_recognition for facial recognition. It seems to perform well on most videos from the training set except on the below videos. Anybody else facing similar issues with other face recognition libraries?
\n\nadhsbajydo\nebeknhudxq\nczmqpxrqoh\ncwqlvzefpg\nbbvgxeczei\navvdgsennp\ncycacemkmt\neoewqcpbgt\ndjvutyvaio\naxwgcsyphv\nesyhwdfnxs\ndhoqofwoxa
", "rawMarkdown": "hey guys, I'm using https://github.com/ageitgey/face_recognition for facial recognition. It seems to perform well on most videos from the training set except on the below videos. Anybody else facing similar issues with other face recognition libraries? \n\n\nadhsbajydo\nebeknhudxq\nczmqpxrqoh\ncwqlvzefpg\nbbvgxeczei\navvdgsennp\ncycacemkmt\neoewqcpbgt\ndjvutyvaio\naxwgcsyphv\nesyhwdfnxs\ndhoqofwoxa", "votes": 4 }, { "id": 753501, "postDate": "2020-02-22T09:37:01.253Z", "content": "Note that facenet mtcnn has detect method. When using it, you can control the probability threshold of the detected faces. So, you can try detection, and if you get no faces at say, 0.6 probability, try lowering the th for this video until you get one or two faces.
", "rawMarkdown": "Note that facenet mtcnn has detect method. When using it, you can control the probability threshold of the detected faces. So, you can try detection, and if you get no faces at say, 0.6 probability, try lowering the th for this video until you get one or two faces.", "votes": 1 }, { "id": 744680, "postDate": "2020-02-13T03:09:58.523Z", "content": "I haven't checked with these specific videos. But I have seen cases where the face detector doesn't return any faces. Your option at that point of time should be to manually predict a 0.5 for the purposes of submission.
", "rawMarkdown": "I haven't checked with these specific videos. But I have seen cases where the face detector doesn't return any faces. Your option at that point of time should be to manually predict a 0.5 for the purposes of submission.", "votes": 1 }, { "id": 744905, "postDate": "2020-02-13T09:37:10.907Z", "content": "I faced the same issue tried many models, and setteled with Dual Shot Face Detector ==> https://github.com/hukkelas/DSFD-Pytorch-Inference
", "rawMarkdown": "I faced the same issue tried many models, and setteled with Dual Shot Face Detector ==> https://github.com/hukkelas/DSFD-Pytorch-Inference\n\n", "votes": 2, "replies": [ { "id": 746019, "postDate": "2020-02-14T14:01:42.083Z", "content": "Always keep in mind that DSD, is slow..., where the one we are using (will not tell which one) is so many times faster and out of ~1200000 (1.2 million) images, failed to detect ~2 faces, and misclassified ~1000, which in total does not make it the best but when it comes to speed and accuracy, it is the best one.
", "rawMarkdown": "Always keep in mind that DSD, is slow..., where the one we are using (will not tell which one) is so many times faster and out of ~1200000 (1.2 million) images, failed to detect ~2 faces, and misclassified ~1000, which in total does not make it the best but when it comes to speed and accuracy, it is the best one.", "votes": 3 }, { "id": 746670, "postDate": "2020-02-15T11:15:20.600Z", "content": "How long does it take to process one video I wonder?
", "rawMarkdown": "How long does it take to process one video I wonder? " }, { "id": 760217, "postDate": "2020-03-01T01:36:00.490Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 753187, "postDate": "2020-02-21T20:33:16.103Z", "content": "Update: I also tried MTCNN, frames extracted from the above images seem to not be recognized even by MTCNN.
", "rawMarkdown": "Update: I also tried MTCNN, frames extracted from the above images seem to not be recognized even by MTCNN. ", "replies": [ { "id": 753194, "postDate": "2020-02-21T20:41:01.423Z", "content": "MTCNN is not great, Starter kernel coming out with LRCN and our face detector implementation soon, please wait a few days.
", "rawMarkdown": "MTCNN is not great, Starter kernel coming out with LRCN and our face detector implementation soon, please wait a few days.", "votes": 4 } ] }, { "id": 745430, "postDate": "2020-02-13T20:05:58.230Z", "content": "Thanks for the suggestions @soviet and @debanga
", "rawMarkdown": "Thanks for the suggestions @soviet and @debanga " }, { "id": 744746, "postDate": "2020-02-13T05:13:08.303Z", "content": "Thanks for the response Akash. Got it, I have been dealing with them in a similar fashion. Can you point me towards other face recognition libraries apart from the one I'm using?
", "rawMarkdown": "Thanks for the response Akash. Got it, I have been dealing with them in a similar fashion. Can you point me towards other face recognition libraries apart from the one I'm using? \n", "replies": [ { "id": 745636, "postDate": "2020-02-14T02:59:17.413Z", "content": "@shivangeetrivedi , I had a look at the github link you shared and it appears that the library can also be used for tagging faces. This is helpful for creating a clear train/val set. I personally am using MTCNN (Tensorflow).
", "rawMarkdown": "@shivangeetrivedi , I had a look at the github link you shared and it appears that the library can also be used for tagging faces. This is helpful for creating a clear train/val set. I personally am using MTCNN (Tensorflow).", "votes": 1 } ] } ], "comments": [ { "id": 744848, "author_name": "Debanga Raj Neog", "author_url": "", "post_date": "2020-02-13T07:42:19.480000", "content": "The ones I tried:
\n\n1) Uses MTCNN (Keras)
\n\nhttps://www.kaggle.com/unkownhihi/starter-kernel-with-cnn-model-ll-lb-0-69235
\n\n2) Uses MTCNN (torch)
\n\nhttps://www.kaggle.com/timesler/facial-recognition-model-in-pytorch
\n\n3) Uses Blazeface (torch)
\n\nhttps://www.kaggle.com/humananalog/inference-demo
", "votes": 4, "replies": [] }, { "id": 753501, "author_name": "Moshel", "author_url": "", "post_date": "2020-02-22T09:37:01.253000", "content": "Note that facenet mtcnn has detect method. When using it, you can control the probability threshold of the detected faces. So, you can try detection, and if you get no faces at say, 0.6 probability, try lowering the th for this video until you get one or two faces.
", "votes": 1, "replies": [] }, { "id": 744680, "author_name": "Akash", "author_url": "", "post_date": "2020-02-13T03:09:58.523000", "content": "I haven't checked with these specific videos. But I have seen cases where the face detector doesn't return any faces. Your option at that point of time should be to manually predict a 0.5 for the purposes of submission.
", "votes": 1, "replies": [] }, { "id": 744905, "author_name": "Soviet", "author_url": "", "post_date": "2020-02-13T09:37:10.907000", "content": "I faced the same issue tried many models, and setteled with Dual Shot Face Detector ==> https://github.com/hukkelas/DSFD-Pytorch-Inference
", "votes": 2, "replies": [ { "id": 746019, "author_name": "Harshit Sheoran", "author_url": "", "post_date": "2020-02-14T14:01:42.083000", "content": "Always keep in mind that DSD, is slow..., where the one we are using (will not tell which one) is so many times faster and out of ~1200000 (1.2 million) images, failed to detect ~2 faces, and misclassified ~1000, which in total does not make it the best but when it comes to speed and accuracy, it is the best one.
", "votes": 3, "replies": [] }, { "id": 746670, "author_name": "yimacs", "author_url": "", "post_date": "2020-02-15T11:15:20.600000", "content": "How long does it take to process one video I wonder?
", "votes": 0, "replies": [] }, { "id": 760217, "author_name": "", "author_url": "", "post_date": "2020-03-01T01:36:00.490000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 753187, "author_name": "Shivangee Trivedi", "author_url": "", "post_date": "2020-02-21T20:33:16.103000", "content": "Update: I also tried MTCNN, frames extracted from the above images seem to not be recognized even by MTCNN.
", "votes": 0, "replies": [ { "id": 753194, "author_name": "Harshit Sheoran", "author_url": "", "post_date": "2020-02-21T20:41:01.423000", "content": "MTCNN is not great, Starter kernel coming out with LRCN and our face detector implementation soon, please wait a few days.
", "votes": 4, "replies": [] } ] }, { "id": 745430, "author_name": "Shivangee Trivedi", "author_url": "", "post_date": "2020-02-13T20:05:58.230000", "content": "Thanks for the suggestions @soviet and @debanga
", "votes": 0, "replies": [] }, { "id": 744746, "author_name": "Shivangee Trivedi", "author_url": "", "post_date": "2020-02-13T05:13:08.303000", "content": "Thanks for the response Akash. Got it, I have been dealing with them in a similar fashion. Can you point me towards other face recognition libraries apart from the one I'm using?
", "votes": 0, "replies": [ { "id": 745636, "author_name": "Akash", "author_url": "", "post_date": "2020-02-14T02:59:17.413000", "content": "@shivangeetrivedi , I had a look at the github link you shared and it appears that the library can also be used for tagging faces. This is helpful for creating a clear train/val set. I personally am using MTCNN (Tensorflow).
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "744848": "The ones I tried:\n\n1) Uses MTCNN (Keras)\n\nhttps://www.kaggle.com/unkownhihi/starter-kernel-with-cnn-model-ll-lb-0-69235\n\n2) Uses MTCNN (torch)\n\nhttps://www.kaggle.com/timesler/facial-recognition-model-in-pytorch\n\n3) Uses Blazeface (torch)\n\nhttps://www.kaggle.com/humananalog/inference-demo", "744424": "hey guys, I'm using https://github.com/ageitgey/face_recognition for facial recognition. It seems to perform well on most videos from the training set except on the below videos. Anybody else facing similar issues with other face recognition libraries? \n\n\nadhsbajydo\nebeknhudxq\nczmqpxrqoh\ncwqlvzefpg\nbbvgxeczei\navvdgsennp\ncycacemkmt\neoewqcpbgt\ndjvutyvaio\naxwgcsyphv\nesyhwdfnxs\ndhoqofwoxa", "753501": "Note that facenet mtcnn has detect method. When using it, you can control the probability threshold of the detected faces. So, you can try detection, and if you get no faces at say, 0.6 probability, try lowering the th for this video until you get one or two faces.", "744680": "I haven't checked with these specific videos. But I have seen cases where the face detector doesn't return any faces. Your option at that point of time should be to manually predict a 0.5 for the purposes of submission.", "744905": "I faced the same issue tried many models, and setteled with Dual Shot Face Detector ==> https://github.com/hukkelas/DSFD-Pytorch-Inference\n\n", "753187": "Update: I also tried MTCNN, frames extracted from the above images seem to not be recognized even by MTCNN. ", "745430": "Thanks for the suggestions @soviet and @debanga ", "744746": "Thanks for the response Akash. Got it, I have been dealing with them in a similar fashion. Can you point me towards other face recognition libraries apart from the one I'm using? \n" } }