Hello community, I'm new to the competition.
\nIn the data description, it's writtent that:
\n\n\n[x/y/z] - The normalized spatial coordinates of the landmark. These are the only columns that will be provided to your submitted model for inference. The MediaPipe model is not fully trained to predict depth so you may wish to ignore the z values.
\n
Does it mean that these are the only columns we should use as input to the model ? Only 3 ?
", "messages": [ { "id": "2186322", "postDate": "03/17/2023 17:54:57", "content": "Hello community, I'm new to the competition.
\nIn the data description, it's writtent that:
\n\n\n[x/y/z] - The normalized spatial coordinates of the landmark. These are the only columns that will be provided to your submitted model for inference. The MediaPipe model is not fully trained to predict depth so you may wish to ignore the z values.
\n
Does it mean that these are the only columns we should use as input to the model ? Only 3 ?
", "rawMarkdown": "Hello community, I'm new to the competition.\n\nIn the data description, it's writtent that:\n>[x/y/z] - The normalized spatial coordinates of the landmark. These are the only columns that will be provided to your submitted model for inference. The MediaPipe model is not fully trained to predict depth so you may wish to ignore the z values. \n\nDoes it mean that these are the only columns we should use as input to the model ? Only 3 ?", "votes": null }, { "id": "2186534", "postDate": "03/17/2023 21:17:37", "content": "My understanding from this that for inference they will provide these columns but we can include a preprocessing layer and get the relevant features t our model as input.
", "rawMarkdown": "My understanding from this that for inference they will provide these columns but we can include a preprocessing layer and get the relevant features t our model as input.", "votes": null }, { "id": "2191227", "postDate": "03/21/2023 20:12:53", "content": "Hi @rayanaay, your submission file will be fed the normalized spatial coordinates of the landmark using the following code load_relevant_data_subset(path). However, that does not mean that these are the only columns you can use as input to your model. In fact, you could create a new tensorflow Module that would make some modification/processing/feature engineering to the x/y/z column before being fed to the ML model. This tf module should have as an input (543, 3) but depending on the pipeline implemented your model could have any input form!
\nMy tips would be to check tf.keras.Layer.layers and tf.Module subclasses.
Hope that helps!
", "rawMarkdown": "Hi @rayanaay, your submission file will be fed the normalized spatial coordinates of the landmark using the following code `load_relevant_data_subset(path)`. However, that does not mean that these are the only columns you can use as input to your model. In fact, you could create a new tensorflow Module that would make some modification/processing/feature engineering to the x/y/z column before being fed to the ML model. This tf module should have as an input (543, 3) but depending on the pipeline implemented your model could have any input form!\nMy tips would be to check [tf.keras.Layer.layers](https://www.tensorflow.org/api_docs/python/tf/keras/layers/Layer) and [tf.Module](https://www.tensorflow.org/api_docs/python/tf/Module) subclasses.\n\nHope that helps!", "votes": null } ], "comments": [ { "id": 2186534, "author_name": "rashasalim", "author_url": "", "post_date": "03/17/2023 21:17:37", "content": "My understanding from this that for inference they will provide these columns but we can include a preprocessing layer and get the relevant features t our model as input.
", "votes": null, "replies": [] }, { "id": 2191227, "author_name": "josephzahar", "author_url": "", "post_date": "03/21/2023 20:12:53", "content": "Hi @rayanaay, your submission file will be fed the normalized spatial coordinates of the landmark using the following code load_relevant_data_subset(path). However, that does not mean that these are the only columns you can use as input to your model. In fact, you could create a new tensorflow Module that would make some modification/processing/feature engineering to the x/y/z column before being fed to the ML model. This tf module should have as an input (543, 3) but depending on the pipeline implemented your model could have any input form!
\nMy tips would be to check tf.keras.Layer.layers and tf.Module subclasses.
Hope that helps!
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2186322": "Hello community, I'm new to the competition.\n\nIn the data description, it's writtent that:\n>[x/y/z] - The normalized spatial coordinates of the landmark. These are the only columns that will be provided to your submitted model for inference. The MediaPipe model is not fully trained to predict depth so you may wish to ignore the z values. \n\nDoes it mean that these are the only columns we should use as input to the model ? Only 3 ?", "2186534": "My understanding from this that for inference they will provide these columns but we can include a preprocessing layer and get the relevant features t our model as input.", "2191227": "Hi @rayanaay, your submission file will be fed the normalized spatial coordinates of the landmark using the following code `load_relevant_data_subset(path)`. However, that does not mean that these are the only columns you can use as input to your model. In fact, you could create a new tensorflow Module that would make some modification/processing/feature engineering to the x/y/z column before being fed to the ML model. This tf module should have as an input (543, 3) but depending on the pipeline implemented your model could have any input form!\nMy tips would be to check [tf.keras.Layer.layers](https://www.tensorflow.org/api_docs/python/tf/keras/layers/Layer) and [tf.Module](https://www.tensorflow.org/api_docs/python/tf/Module) subclasses.\n\nHope that helps!" }, "source": "meta" }