![\"metrics\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17954496%2F8a035850422c1c61dadd076698f2531d%2Fmetrics2.png?generation=1762898881801102&alt=media\")
Hello everyone, I would like to show the results for the base 10 epochs. Please write in the comments if you would be interested in reviewing a notebook with information about the models and their scores. Also let me know which models can be viewed and compared more.
\n
Hello everyone, I would like to show the results for the base 10 epochs. Please write in the comments if you would be interested in reviewing a notebook with information about the models and their scores. Also let me know which models can be viewed and compared more.
\n
From these graph, I feel the DINOv2 produce best performance. What do you comment on this ?
", "rawMarkdown": "From these graph, I feel the DINOv2 produce best performance. What do you comment on this ?", "votes": null }, { "id": "3361719", "postDate": "12/03/2025 18:42:11", "content": "According to the metrics, DINOv2 stands out as the superior model, demonstrating the highest and most stable performance across both Validation IoU and F1 scores. DeepLabV3 follows as a strong contender, achieving the lowest training loss and competitive peak F1 scores despite showing slightly higher variance. I think the highest scores are currently achieved based on DINOv2.
", "rawMarkdown": "According to the metrics, DINOv2 stands out as the superior model, demonstrating the highest and most stable performance across both Validation IoU and F1 scores. DeepLabV3 follows as a strong contender, achieving the lowest training loss and competitive peak F1 scores despite showing slightly higher variance. I think the highest scores are currently achieved based on DINOv2.", "votes": null }, { "id": "3367888", "postDate": "12/08/2025 19:46:25", "content": "I tried other backbone like MobileNet, ResNet, EfficientNet and I agree that I had the best performance using dino
", "rawMarkdown": "I tried other backbone like MobileNet, ResNet, EfficientNet and I agree that I had the best performance using dino", "votes": null }, { "id": "3377040", "postDate": "12/15/2025 15:41:26", "content": "my question is about DINO , the model indeed worked very welll , but during submission time kaggle wont accept as loading the trained model required external dependencies which required internet , so how did u do it??
", "rawMarkdown": "my question is about DINO , the model indeed worked very welll , but during submission time kaggle wont accept as loading the trained model required external dependencies which required internet , so how did u do it??", "votes": null }, { "id": "3377130", "postDate": "12/15/2025 19:34:10", "content": "On the top left of your notebook click \"settings\" and from the dropdown select \"turn off internet\", then from the panel on the right click on \"add input\" and search for dinoV2, in fine select the version you will need. Happy training!
", "rawMarkdown": "On the top left of your notebook click \"settings\" and from the dropdown select \"turn off internet\", then from the panel on the right click on \"add input\" and search for dinoV2, in fine select the version you will need. Happy training!", "votes": null } ], "comments": [ { "id": 3354316, "author_name": "yxyyxy", "author_url": "", "post_date": "11/30/2025 08:51:46", "content": "From these graph, I feel the DINOv2 produce best performance. What do you comment on this ?
", "votes": null, "replies": [ { "id": 3367888, "author_name": "mohamedidrissighalmi", "author_url": "", "post_date": "12/08/2025 19:46:25", "content": "I tried other backbone like MobileNet, ResNet, EfficientNet and I agree that I had the best performance using dino
", "votes": null, "replies": [] } ] }, { "id": 3361719, "author_name": "musapeker", "author_url": "", "post_date": "12/03/2025 18:42:11", "content": "According to the metrics, DINOv2 stands out as the superior model, demonstrating the highest and most stable performance across both Validation IoU and F1 scores. DeepLabV3 follows as a strong contender, achieving the lowest training loss and competitive peak F1 scores despite showing slightly higher variance. I think the highest scores are currently achieved based on DINOv2.
", "votes": null, "replies": [] }, { "id": 3377040, "author_name": "qudsiyasiddique", "author_url": "", "post_date": "12/15/2025 15:41:26", "content": "my question is about DINO , the model indeed worked very welll , but during submission time kaggle wont accept as loading the trained model required external dependencies which required internet , so how did u do it??
", "votes": null, "replies": [ { "id": 3377130, "author_name": "mohamedidrissighalmi", "author_url": "", "post_date": "12/15/2025 19:34:10", "content": "On the top left of your notebook click \"settings\" and from the dropdown select \"turn off internet\", then from the panel on the right click on \"add input\" and search for dinoV2, in fine select the version you will need. Happy training!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3319591": "Hello everyone, I would like to show the results for the base 10 epochs. Please write in the comments if you would be interested in reviewing a notebook with information about the models and their scores. Also let me know which models can be viewed and compared more.\n\n", "3354316": "From these graph, I feel the DINOv2 produce best performance. What do you comment on this ?", "3361719": "According to the metrics, DINOv2 stands out as the superior model, demonstrating the highest and most stable performance across both Validation IoU and F1 scores. DeepLabV3 follows as a strong contender, achieving the lowest training loss and competitive peak F1 scores despite showing slightly higher variance. I think the highest scores are currently achieved based on DINOv2.", "3367888": "I tried other backbone like MobileNet, ResNet, EfficientNet and I agree that I had the best performance using dino", "3377040": "my question is about DINO , the model indeed worked very welll , but during submission time kaggle wont accept as loading the trained model required external dependencies which required internet , so how did u do it??", "3377130": "On the top left of your notebook click \"settings\" and from the dropdown select \"turn off internet\", then from the panel on the right click on \"add input\" and search for dinoV2, in fine select the version you will need. Happy training!" }, "source": "meta" }