Hi, I would like to perform sample augmentation by duplication of rows in the training set, how can I add new rows to the dataset table? \nI duplicated images and labels locally then loaded with train_table = tlc.Table.from_url', however row_count does not increase. How can I register this change in the table?
Hi, I would like to perform sample augmentation by duplication of rows in the training set, how can I add new rows to the dataset table? \nI duplicated images and labels locally then loaded with train_table = tlc.Table.from_url', however row_count does not increase. How can I register this change in the table?
Hi @marcobuiani. If you would like certain images to appear more often in training, I recommend using Sample Weights instead. In the Dashboard, edit the WEIGHT column value for the rows you would like to appear more often, for example to a value of 2. Then commit that change to create a Table Revision. In the YOLO integration, use the new Table and set sampling_weights=True in your Settings object and pass that to model.train(). The samples with weight 2 will appear twice as often as those with weight 1.
Similarly, if you would like to ignore certain samples, you can set their WEIGHT to 0.
To read more about the YOLO Settings, see https://github.com/3lc-ai/3lc-ultralytics?tab=readme-ov-file#3lc-settings\nThis article details how to edit a number of rows at once in the Dashboard: https://docs.3lc.ai/3lc/latest/how-to/basics/create-column.html#edit-an-editable-column
", "rawMarkdown": "Hi @marcobuiani. If you would like certain images to appear more often in training, I recommend using Sample Weights instead. In the Dashboard, edit the `WEIGHT` column value for the rows you would like to appear more often, for example to a value of 2. Then commit that change to create a Table Revision. In the YOLO integration, use the new Table and set `sampling_weights=True` in your Settings object and pass that to `model.train()`. The samples with weight 2 will appear twice as often as those with weight 1.\n\nSimilarly, if you would like to ignore certain samples, you can set their `WEIGHT` to 0.\n \nTo read more about the YOLO Settings, see https://github.com/3lc-ai/3lc-ultralytics?tab=readme-ov-file#3lc-settings\nThis article details how to edit a number of rows at once in the Dashboard: https://docs.3lc.ai/3lc/latest/how-to/basics/create-column.html#edit-an-editable-column", "votes": null }, { "id": "3350549", "postDate": "11/27/2025 16:11:53", "content": "Hi @fredmell thanks for clarifying, now I understand I must use sampling_weights=True for duplication. However I wanted to do something different: I wanted to add samples for background, new images (from train crops) without labels. But if I add them locally and register a new table I will lose the labels and BB modifications I did on my current tables. The only way I can think of is to export my current table locally, add the images and register the new one.
You can create a new table with only the new images and then join this new table with another table (probably your latest revision).
\njoined_table = tlc.Table.join_tables([initial_table, new_data_table], table_name=\"added-more-data\")
Learn more about adding new data at this example notebook https://github.com/3lc-ai/3lc-examples/blob/main/tutorials/2-modify-tables/add-new-data.ipynb
", "rawMarkdown": "You can create a new table with only the new images and then join this new table with another table (probably your latest revision). \n\n`joined_table = tlc.Table.join_tables([initial_table, new_data_table], table_name=\"added-more-data\")`\n\nLearn more about adding new data at this example notebook https://github.com/3lc-ai/3lc-examples/blob/main/tutorials/2-modify-tables/add-new-data.ipynb", "votes": null } ], "comments": [ { "id": 3350380, "author_name": "fredmell", "author_url": "", "post_date": "11/27/2025 14:30:09", "content": "Hi @marcobuiani. If you would like certain images to appear more often in training, I recommend using Sample Weights instead. In the Dashboard, edit the WEIGHT column value for the rows you would like to appear more often, for example to a value of 2. Then commit that change to create a Table Revision. In the YOLO integration, use the new Table and set sampling_weights=True in your Settings object and pass that to model.train(). The samples with weight 2 will appear twice as often as those with weight 1.
Similarly, if you would like to ignore certain samples, you can set their WEIGHT to 0.
To read more about the YOLO Settings, see https://github.com/3lc-ai/3lc-ultralytics?tab=readme-ov-file#3lc-settings\nThis article details how to edit a number of rows at once in the Dashboard: https://docs.3lc.ai/3lc/latest/how-to/basics/create-column.html#edit-an-editable-column
", "votes": null, "replies": [ { "id": 3350549, "author_name": "marcobuiani", "author_url": "", "post_date": "11/27/2025 16:11:53", "content": "Hi @fredmell thanks for clarifying, now I understand I must use sampling_weights=True for duplication. However I wanted to do something different: I wanted to add samples for background, new images (from train crops) without labels. But if I add them locally and register a new table I will lose the labels and BB modifications I did on my current tables. The only way I can think of is to export my current table locally, add the images and register the new one.
You can create a new table with only the new images and then join this new table with another table (probably your latest revision).
\njoined_table = tlc.Table.join_tables([initial_table, new_data_table], table_name=\"added-more-data\")
Learn more about adding new data at this example notebook https://github.com/3lc-ai/3lc-examples/blob/main/tutorials/2-modify-tables/add-new-data.ipynb
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3349758": "Hi, I would like to perform sample augmentation by duplication of rows in the training set, how can I add new rows to the dataset table? \nI duplicated images and labels locally then loaded with `train_table = tlc.Table.from_url'`, however `row_count` does not increase. How can I register this change in the table?", "3350380": "Hi @marcobuiani. If you would like certain images to appear more often in training, I recommend using Sample Weights instead. In the Dashboard, edit the `WEIGHT` column value for the rows you would like to appear more often, for example to a value of 2. Then commit that change to create a Table Revision. In the YOLO integration, use the new Table and set `sampling_weights=True` in your Settings object and pass that to `model.train()`. The samples with weight 2 will appear twice as often as those with weight 1.\n\nSimilarly, if you would like to ignore certain samples, you can set their `WEIGHT` to 0.\n \nTo read more about the YOLO Settings, see https://github.com/3lc-ai/3lc-ultralytics?tab=readme-ov-file#3lc-settings\nThis article details how to edit a number of rows at once in the Dashboard: https://docs.3lc.ai/3lc/latest/how-to/basics/create-column.html#edit-an-editable-column", "3350549": "Hi @fredmell thanks for clarifying, now I understand I must use `sampling_weights=True` for duplication. However I wanted to do something different: I wanted to add samples for background, new images (from train crops) without labels. But if I add them locally and register a new table I will lose the labels and BB modifications I did on my current tables. The only way I can think of is to export my current table locally, add the images and register the new one.", "3350716": "You can create a new table with only the new images and then join this new table with another table (probably your latest revision). \n\n`joined_table = tlc.Table.join_tables([initial_table, new_data_table], table_name=\"added-more-data\")`\n\nLearn more about adding new data at this example notebook https://github.com/3lc-ai/3lc-examples/blob/main/tutorials/2-modify-tables/add-new-data.ipynb" }, "source": "meta" }