![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2Ffaf748d058abd58aa27a90408d57fca7%2FScreen%20Shot%202020-07-29%20at%209.43.21%20AM.png?generation=1596041019402634&alt=media\")
For those using TensorFlow, I have a GitHub repository here that contains ResNeXt and ResNeSt models with pretrained weights. I know many people are sticking with EfficientNets, but more ensembling, having a variety of models may help.
", "messages": [ { "id": "950783", "postDate": "07/29/2020 16:00:00", "content": "For those using TensorFlow, I have a GitHub repository here that contains ResNeXt and ResNeSt models with pretrained weights. I know many people are sticking with EfficientNets, but more ensembling, having a variety of models may help.
", "rawMarkdown": "For those using TensorFlow, I have a GitHub repository [here](https://github.com/RichardXiao13/TensorFlow-ResNets) that contains ResNeXt and ResNeSt models with pretrained weights. I know many people are sticking with EfficientNets, but more ensembling, having a variety of models may help.", "votes": null }, { "id": "950829", "postDate": "07/29/2020 16:45:40", "content": "Hi Richard, thanks for the link. There is a huge repository of all TF models here. It's Qubvel's classification models. He has every model.
\n\n
Hi Richard, I personally other models such as ResneXt are performing poor wrt Efficient Nets. It is a good idea to ensemble different models for final submission. I looked at the winning solution of a previous Skin cancer detection challenge. The winning solution had ensemble of 10 different models. Out of which, 3 were Resnext with different hyper parameters and 2 DenseNet models.
", "rawMarkdown": "Hi Richard, I personally other models such as ResneXt are performing poor wrt Efficient Nets. It is a good idea to ensemble different models for final submission. I looked at the winning solution of a previous Skin cancer detection challenge. The winning solution had ensemble of 10 different models. Out of which, 3 were Resnext with different hyper parameters and 2 DenseNet models.", "votes": null } ], "comments": [ { "id": 950829, "author_name": "cdeotte", "author_url": "", "post_date": "07/29/2020 16:45:40", "content": "Hi Richard, thanks for the link. There is a huge repository of all TF models here. It's Qubvel's classification models. He has every model.
\n\n
Hi Richard, I personally other models such as ResneXt are performing poor wrt Efficient Nets. It is a good idea to ensemble different models for final submission. I looked at the winning solution of a previous Skin cancer detection challenge. The winning solution had ensemble of 10 different models. Out of which, 3 were Resnext with different hyper parameters and 2 DenseNet models.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "950783": "For those using TensorFlow, I have a GitHub repository [here](https://github.com/RichardXiao13/TensorFlow-ResNets) that contains ResNeXt and ResNeSt models with pretrained weights. I know many people are sticking with EfficientNets, but more ensembling, having a variety of models may help.", "950829": "Hi Richard, thanks for the link. There is a huge repository of all TF models [here][1]. It's Qubvel's classification models. He has every model.\n\n\n\n\n[1]: https://github.com/qubvel/classification_models", "958516": "Hi Richard, I personally other models such as ResneXt are performing poor wrt Efficient Nets. It is a good idea to ensemble different models for final submission. I looked at the winning solution of a previous Skin cancer detection challenge. The winning solution had ensemble of 10 different models. Out of which, 3 were Resnext with different hyper parameters and 2 DenseNet models." }, "source": "meta" }