Hello! \nI am trying to find the best way to run pre-trained YOLOv4 to get predictions. I found some public kernels based on PyTorch, but:\n1) I am usually using Keras and TensorFlow, so PyTorch solutions are some kind of \"black box\" for me.\n2) Sometimes they use GitHub projects under GPT license and I try to stick on competition rules.
\n\nI found that you can use YOLOv4 from OpenCV 3.4 branch: https://github.com/opencv/opencv/pull/17185\nIt's currently not in the master branch, so, if I understand it correctly, I can`t just use pip to install it.
\n\nIs it possible to install this particular version into my Kaggle Kernel?
", "messages": [ { "id": "950266", "postDate": "07/29/2020 09:58:46", "content": "Hello! \nI am trying to find the best way to run pre-trained YOLOv4 to get predictions. I found some public kernels based on PyTorch, but:\n1) I am usually using Keras and TensorFlow, so PyTorch solutions are some kind of \"black box\" for me.\n2) Sometimes they use GitHub projects under GPT license and I try to stick on competition rules.
\n\nI found that you can use YOLOv4 from OpenCV 3.4 branch: https://github.com/opencv/opencv/pull/17185\nIt's currently not in the master branch, so, if I understand it correctly, I can`t just use pip to install it.
\n\nIs it possible to install this particular version into my Kaggle Kernel?
", "rawMarkdown": "Hello! \nI am trying to find the best way to run pre-trained YOLOv4 to get predictions. I found some public kernels based on PyTorch, but:\n1) I am usually using Keras and TensorFlow, so PyTorch solutions are some kind of \"black box\" for me.\n2) Sometimes they use GitHub projects under GPT license and I try to stick on competition rules.\n\nI found that you can use YOLOv4 from OpenCV 3.4 branch: https://github.com/opencv/opencv/pull/17185\nIt's currently not in the master branch, so, if I understand it correctly, I can`t just use pip to install it.\n\nIs it possible to install this particular version into my Kaggle Kernel?", "votes": null }, { "id": "950525", "postDate": "07/29/2020 12:57:30", "content": "@gooogr OpenCV has released it's new version with support of Darknet's YOLOv4 and tiny YOLOv4. You can find it's details here. Currently Kaggle uses OpenCV with it's Python wrapper (opencv-python) which is not updated with YOLOv4 support.
\n\nIf you want to use pre-trained YOLOv4 weights for inference then you can follow below notebook-\nusing Darknet command for inference
", "rawMarkdown": "gooogr OpenCV has released it's new version with support of Darknet's YOLOv4 and tiny YOLOv4. You can find it's details [here.](https://github.com/opencv/opencv/tree/4.4.0) Currently Kaggle uses OpenCV with it's Python wrapper (opencv-python) which is not updated with YOLOv4 support.\n\nIf you want to use pre-trained YOLOv4 weights for inference then you can follow below notebook-\n[using Darknet command for inference](https://www.kaggle.com/pabloberhauser/yolov4-darknet-inference)", "votes": null }, { "id": "950854", "postDate": "07/29/2020 17:13:03", "content": "Thank you! I seem to be a little confused about the releases of the versions.\nYes, I also found that notebook and wrote my kernel with YOLO Darknet implementation. It's not such convenient as using OpenCV (3 lines of code and your model is ready ), but friendly than pablOberhauser's. \nIn case someone find this thread, here is the link
", "rawMarkdown": "Thank you! I seem to be a little confused about the releases of the versions.\nYes, I also found that notebook and wrote my kernel with YOLO Darknet implementation. It's not such convenient as using OpenCV (3 lines of code and your model is ready ), but friendly than pablOberhauser's. \nIn case someone find this thread, [here is the link](https://www.kaggle.com/gooogr/gwd-predicts-with-yolov4-and-darknet)", "votes": null }, { "id": "951804", "postDate": "07/30/2020 12:27:23", "content": "@gooogr I am glad that you are finally able to proceed. So what is you score?
", "rawMarkdown": "gooogr I am glad that you are finally able to proceed. So what is you score?", "votes": null } ], "comments": [ { "id": 950525, "author_name": "iamprateek", "author_url": "", "post_date": "07/29/2020 12:57:30", "content": "@gooogr OpenCV has released it's new version with support of Darknet's YOLOv4 and tiny YOLOv4. You can find it's details here. Currently Kaggle uses OpenCV with it's Python wrapper (opencv-python) which is not updated with YOLOv4 support.
\n\nIf you want to use pre-trained YOLOv4 weights for inference then you can follow below notebook-\nusing Darknet command for inference
", "votes": null, "replies": [ { "id": 950854, "author_name": "gooogr", "author_url": "", "post_date": "07/29/2020 17:13:03", "content": "Thank you! I seem to be a little confused about the releases of the versions.\nYes, I also found that notebook and wrote my kernel with YOLO Darknet implementation. It's not such convenient as using OpenCV (3 lines of code and your model is ready ), but friendly than pablOberhauser's. \nIn case someone find this thread, here is the link
", "votes": null, "replies": [] }, { "id": 951804, "author_name": "iamprateek", "author_url": "", "post_date": "07/30/2020 12:27:23", "content": "@gooogr I am glad that you are finally able to proceed. So what is you score?
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "950266": "Hello! \nI am trying to find the best way to run pre-trained YOLOv4 to get predictions. I found some public kernels based on PyTorch, but:\n1) I am usually using Keras and TensorFlow, so PyTorch solutions are some kind of \"black box\" for me.\n2) Sometimes they use GitHub projects under GPT license and I try to stick on competition rules.\n\nI found that you can use YOLOv4 from OpenCV 3.4 branch: https://github.com/opencv/opencv/pull/17185\nIt's currently not in the master branch, so, if I understand it correctly, I can`t just use pip to install it.\n\nIs it possible to install this particular version into my Kaggle Kernel?", "950525": "gooogr OpenCV has released it's new version with support of Darknet's YOLOv4 and tiny YOLOv4. You can find it's details [here.](https://github.com/opencv/opencv/tree/4.4.0) Currently Kaggle uses OpenCV with it's Python wrapper (opencv-python) which is not updated with YOLOv4 support.\n\nIf you want to use pre-trained YOLOv4 weights for inference then you can follow below notebook-\n[using Darknet command for inference](https://www.kaggle.com/pabloberhauser/yolov4-darknet-inference)", "950854": "Thank you! I seem to be a little confused about the releases of the versions.\nYes, I also found that notebook and wrote my kernel with YOLO Darknet implementation. It's not such convenient as using OpenCV (3 lines of code and your model is ready ), but friendly than pablOberhauser's. \nIn case someone find this thread, [here is the link](https://www.kaggle.com/gooogr/gwd-predicts-with-yolov4-and-darknet)", "951804": "gooogr I am glad that you are finally able to proceed. So what is you score?" }, "source": "meta" }