\nThis is the model
![\"Model's](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F13549016%2Fe6216f66e774fa919fb6521f53b2d07b%2FScreenshot%202025-08-09%20104729.png?generation=1754725671165057&alt=media\")
Hello Peeps,
\nI was kind of bothered by the high-dimensionality and the sparisty of the tof sensor data. And I didn't encounter a solution that addressed this concern. So I built a convolutional autoencoder to make the data more NN friendly, and also to be able to train models that are much smaller in size. I am currently experimenting with different ways of post processing the output of the encoder (which is a 5x5x16 tensor). Nonetheless, I have made the model's weights public along with a baseline LSTM that uses the compressed tof data.
\nThis is the model
Hello Peeps,
\nI was kind of bothered by the high-dimensionality and the sparisty of the tof sensor data. And I didn't encounter a solution that addressed this concern. So I built a convolutional autoencoder to make the data more NN friendly, and also to be able to train models that are much smaller in size. I am currently experimenting with different ways of post processing the output of the encoder (which is a 5x5x16 tensor). Nonetheless, I have made the model's weights public along with a baseline LSTM that uses the compressed tof data.
\nThis is the model
https://www.kaggle.com/datasets/moetasimrady/tof-autoencoder; This dataset houses the model's weights.
\nAnd this is the model's training data: ![\"Model_training.jpg\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F13549016%2F788504447506025b5a902550421d4585%2Fautoencoder%20training.png?generation=1754725564321130&alt=media\")
And this is the NB of the baseline LSTM with the compression preprocessing steps Baseline LSTM with compressed tof
", "rawMarkdown": "[https://www.kaggle.com/datasets/moetasimrady/tof-autoencoder](Tof-autoencoder); This dataset houses the model's weights.\nAnd this is the model's training data: \n\nAnd this is the NB of the baseline LSTM with the compression preprocessing steps [Baseline LSTM with compressed tof](https://www.kaggle.com/code/moetasimrady/cmi-lstm-classifier-with-tof-and-thm)", "votes": null } ], "comments": [ { "id": 3266481, "author_name": "moetasimrady", "author_url": "", "post_date": "08/09/2025 07:46:31", "content": "https://www.kaggle.com/datasets/moetasimrady/tof-autoencoder; This dataset houses the model's weights.
\nAnd this is the model's training data:
And this is the NB of the baseline LSTM with the compression preprocessing steps Baseline LSTM with compressed tof
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3266480": "Hello Peeps,\n\nI was kind of bothered by the high-dimensionality and the sparisty of the tof sensor data. And I didn't encounter a solution that addressed this concern. So I built a convolutional autoencoder to make the data more NN friendly, and also to be able to train models that are much smaller in size. I am currently experimenting with different ways of post processing the output of the encoder (which is a 5x5x16 tensor). Nonetheless, I have made the model's weights public along with a baseline LSTM that uses the compressed tof data. \nThis is the model ", "3266481": "[https://www.kaggle.com/datasets/moetasimrady/tof-autoencoder](Tof-autoencoder); This dataset houses the model's weights.\nAnd this is the model's training data: \n\nAnd this is the NB of the baseline LSTM with the compression preprocessing steps [Baseline LSTM with compressed tof](https://www.kaggle.com/code/moetasimrady/cmi-lstm-classifier-with-tof-and-thm)" }, "source": "meta" }