{ "id": 409437, "title": "Some ideas for competitions beginner", "url": "/competitions/google-research-identify-contrails-reduce-global-warming/discussion/409437", "author_name": "Dewei Chen", "post_date": "2023-05-11T02:47:10.447000", "votes": 41, "comment_count": 12, "views": 0, "content": "

1. Do some pre-processing

\n

Convert .npy to image

\n
 numpy  np\nnp.load(FILE_NAME)\n
\n

Using Kalman Filter or some others method for estimate the trajectory (tracking)

\n
 :\n     ():\n        self.state = np.array(initial_state, dtype=np.float32)\n        self.process_noise = process_noise\n        self.measurement_noise = measurement_noise\n        self.error_covariance = np.eye(, dtype=np.float32)\n        self.transition_matrix = np.array([[, , , ], \n                                           [, , , ], \n                                           [, , , ], \n                                           [, , , ]], dtype=np.float32)\n        self.measurement_matrix = np.array([[, , , ], \n                                            [, , , ]], dtype=np.float32)\n\n     ():\n        self.state = np.dot(self.transition_matrix, self.state)\n        self.error_covariance = np.dot(np.dot(self.transition_matrix, self.error_covariance), \n                                       self.transition_matrix.T) + self.process_noise\n         self.state[:]\n\n     ():\n        kalman_gain = np.dot(np.dot(self.error_covariance, self.measurement_matrix.T), \n                             np.linalg.inv(np.dot(np.dot(self.measurement_matrix, self.error_covariance), \n                                                  self.measurement_matrix.T) + self.measurement_noise))\n        self.state = self.state + np.dot(kalman_gain, (measurement - np.dot(self.measurement_matrix, self.state)))\n        self.error_covariance = self.error_covariance - np.dot(np.dot(kalman_gain, self.measurement_matrix), self.error_covariance)\n         self. State[:]\n\n\n
\n

2. Using some image classification/detection model Backbone, and implement your own model.

\n

Yolov8: https://github.com/ultralytics/ultralytics
\nUnet: https://github.com/topics/unet
\netc…..

\n

3. You may need a POWERFUL GPU for traing.

", "messages": [ { "id": 2254468, "postDate": "2023-05-11T02:47:10.447Z", "content": "

1. Do some pre-processing

\n

Convert .npy to image

\n
 numpy  np\nnp.load(FILE_NAME)\n
\n

Using Kalman Filter or some others method for estimate the trajectory (tracking)

\n
 :\n     ():\n        self.state = np.array(initial_state, dtype=np.float32)\n        self.process_noise = process_noise\n        self.measurement_noise = measurement_noise\n        self.error_covariance = np.eye(, dtype=np.float32)\n        self.transition_matrix = np.array([[, , , ], \n                                           [, , , ], \n                                           [, , , ], \n                                           [, , , ]], dtype=np.float32)\n        self.measurement_matrix = np.array([[, , , ], \n                                            [, , , ]], dtype=np.float32)\n\n     ():\n        self.state = np.dot(self.transition_matrix, self.state)\n        self.error_covariance = np.dot(np.dot(self.transition_matrix, self.error_covariance), \n                                       self.transition_matrix.T) + self.process_noise\n         self.state[:]\n\n     ():\n        kalman_gain = np.dot(np.dot(self.error_covariance, self.measurement_matrix.T), \n                             np.linalg.inv(np.dot(np.dot(self.measurement_matrix, self.error_covariance), \n                                                  self.measurement_matrix.T) + self.measurement_noise))\n        self.state = self.state + np.dot(kalman_gain, (measurement - np.dot(self.measurement_matrix, self.state)))\n        self.error_covariance = self.error_covariance - np.dot(np.dot(kalman_gain, self.measurement_matrix), self.error_covariance)\n         self. State[:]\n\n\n
\n

2. Using some image classification/detection model Backbone, and implement your own model.

\n

Yolov8: https://github.com/ultralytics/ultralytics
\nUnet: https://github.com/topics/unet
\netc…..

\n

3. You may need a POWERFUL GPU for traing.

", "rawMarkdown": "## 1. Do some pre-processing\n\n\nConvert .npy to image\n```python\nimport numpy as np\nnp.load(FILE_NAME)\n```\n\nUsing Kalman Filter or some others method for estimate the trajectory (tracking)\n```python\nclass KalmanFilterTracker:\n def __init__(self, initial_state, process_noise, measurement_noise):\n self.state = np.array(initial_state, dtype=np.float32)\n self.process_noise = process_noise\n self.measurement_noise = measurement_noise\n self.error_covariance = np.eye(4, dtype=np.float32)\n self.transition_matrix = np.array([[1, 0, 1, 0], \n [0, 1, 0, 1], \n [0, 0, 1, 0], \n [0, 0, 0, 1]], dtype=np.float32)\n self.measurement_matrix = np.array([[1, 0, 0, 0], \n [0, 1, 0, 0]], dtype=np.float32)\n\n def predict(self):\n self.state = np.dot(self.transition_matrix, self.state)\n self.error_covariance = np.dot(np.dot(self.transition_matrix, self.error_covariance), \n self.transition_matrix.T) + self.process_noise\n return self.state[:2]\n\n def correct(self, measurement):\n kalman_gain = np.dot(np.dot(self.error_covariance, self.measurement_matrix.T), \n np.linalg.inv(np.dot(np.dot(self.measurement_matrix, self.error_covariance), \n self.measurement_matrix.T) + self.measurement_noise))\n self.state = self.state + np.dot(kalman_gain, (measurement - np.dot(self.measurement_matrix, self.state)))\n self.error_covariance = self.error_covariance - np.dot(np.dot(kalman_gain, self.measurement_matrix), self.error_covariance)\n return self. State[:2]\n\n# Perform object detection on the current frame to obtain the position of a person, such as using object detection method. Assuming we have detected the position of a person, we can update the Kalman filter to obtain the predicted and corrected position\n```\n\n\n## 2. Using some image classification/detection model Backbone, and implement your own model.\nYolov8: https://github.com/ultralytics/ultralytics\nUnet: https://github.com/topics/unet\netc.....\n\n\n## 3. You may need a POWERFUL GPU for traing.", "votes": 40 }, { "id": 2259514, "postDate": "2023-05-15T04:09:28.710Z", "content": "

is there any notebook available where the kalman filter is implemented on images as such. Link please if found. Upvoted , great discussion thread.

", "rawMarkdown": "is there any notebook available where the kalman filter is implemented on images as such. Link please if found. Upvoted , great discussion thread.", "votes": 1, "replies": [ { "id": 2259884, "postDate": "2023-05-15T09:51:44.573Z", "content": "

Here's a GitHub repo: Kalman-Filter-Image

\n

Let me know if this helps :)

", "rawMarkdown": "Here's a GitHub repo: [Kalman-Filter-Image](https://github.com/Team-Neighborhood/Kalman-Filter-Image)\n\nLet me know if this helps :)", "votes": 4, "replies": [ { "id": 2259896, "postDate": "2023-05-15T10:01:14.077Z", "content": "

Nice project, thanks for forward the link :)

", "rawMarkdown": "Nice project, thanks for forward the link :)", "votes": 1 }, { "id": 2260143, "postDate": "2023-05-15T13:39:16.860Z", "content": "

Thanks! Really helpful!

", "rawMarkdown": "Thanks! Really helpful!" } ] } ] }, { "id": 2259125, "postDate": "2023-05-14T17:42:34.693Z", "content": "

Thank you! This is very useful information.

", "rawMarkdown": "Thank you! This is very useful information. ", "votes": 1 }, { "id": 2254488, "postDate": "2023-05-11T03:38:21.333Z", "content": "

Informative, thanks for share.

", "rawMarkdown": "Informative, thanks for share.", "votes": 1, "replies": [ { "id": 2255839, "postDate": "2023-05-12T03:21:01.927Z", "content": "

You're welcome, i'll go on.

", "rawMarkdown": "You're welcome, i'll go on.", "votes": 1 } ] }, { "id": 2257061, "postDate": "2023-05-13T01:22:45.870Z", "content": "

Thanks! Could you please explain Kalman Filtering a bit more or provide resources for it?

", "rawMarkdown": "Thanks! Could you please explain Kalman Filtering a bit more or provide resources for it?", "votes": 2 }, { "id": 2327293, "postDate": "2023-07-02T20:31:39.637Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2273859, "postDate": "2023-05-25T13:10:46.743Z", "content": "

Thanks for sharing!

", "rawMarkdown": "Thanks for sharing!", "votes": 1 }, { "id": 2265126, "postDate": "2023-05-19T02:17:26.397Z", "content": "

Thanks! Good suggestions.

", "rawMarkdown": "Thanks! Good suggestions.", "votes": 1 }, { "id": 2254510, "postDate": "2023-05-11T03:54:38.507Z", "content": "

Thanks a lot!

", "rawMarkdown": "Thanks a lot!", "votes": 1 } ], "comments": [ { "id": 2259514, "author_name": "Arun Munagala", "author_url": "", "post_date": "2023-05-15T04:09:28.710000", "content": "

is there any notebook available where the kalman filter is implemented on images as such. Link please if found. Upvoted , great discussion thread.

", "votes": 1, "replies": [ { "id": 2259884, "author_name": "Aryan Garg", "author_url": "", "post_date": "2023-05-15T09:51:44.573000", "content": "

Here's a GitHub repo: Kalman-Filter-Image

\n

Let me know if this helps :)

", "votes": 4, "replies": [ { "id": 2259896, "author_name": "Dewei Chen", "author_url": "", "post_date": "2023-05-15T10:01:14.077000", "content": "

Nice project, thanks for forward the link :)

", "votes": 1, "replies": [] }, { "id": 2260143, "author_name": "Arun Munagala", "author_url": "", "post_date": "2023-05-15T13:39:16.860000", "content": "

Thanks! Really helpful!

", "votes": 0, "replies": [] } ] } ] }, { "id": 2259125, "author_name": "Pavel Shunkevich", "author_url": "", "post_date": "2023-05-14T17:42:34.693000", "content": "

Thank you! This is very useful information.

", "votes": 1, "replies": [] }, { "id": 2254488, "author_name": "MD. RIPON MIAH", "author_url": "", "post_date": "2023-05-11T03:38:21.333000", "content": "

Informative, thanks for share.

", "votes": 1, "replies": [ { "id": 2255839, "author_name": "Dewei Chen", "author_url": "", "post_date": "2023-05-12T03:21:01.927000", "content": "

You're welcome, i'll go on.

", "votes": 1, "replies": [] } ] }, { "id": 2257061, "author_name": "Aryan Garg", "author_url": "", "post_date": "2023-05-13T01:22:45.870000", "content": "

Thanks! Could you please explain Kalman Filtering a bit more or provide resources for it?

", "votes": 2, "replies": [] }, { "id": 2327293, "author_name": "", "author_url": "", "post_date": "2023-07-02T20:31:39.637000", "content": "", "votes": 0, "replies": [] }, { "id": 2273859, "author_name": "Manuel Carita", "author_url": "", "post_date": "2023-05-25T13:10:46.743000", "content": "

Thanks for sharing!

", "votes": 1, "replies": [] }, { "id": 2265126, "author_name": "Harrison", "author_url": "", "post_date": "2023-05-19T02:17:26.397000", "content": "

Thanks! Good suggestions.

", "votes": 1, "replies": [] }, { "id": 2254510, "author_name": "Joshua Adrian Cahyono", "author_url": "", "post_date": "2023-05-11T03:54:38.507000", "content": "

Thanks a lot!

", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "2254468": "## 1. Do some pre-processing\n\n\nConvert .npy to image\n```python\nimport numpy as np\nnp.load(FILE_NAME)\n```\n\nUsing Kalman Filter or some others method for estimate the trajectory (tracking)\n```python\nclass KalmanFilterTracker:\n def __init__(self, initial_state, process_noise, measurement_noise):\n self.state = np.array(initial_state, dtype=np.float32)\n self.process_noise = process_noise\n self.measurement_noise = measurement_noise\n self.error_covariance = np.eye(4, dtype=np.float32)\n self.transition_matrix = np.array([[1, 0, 1, 0], \n [0, 1, 0, 1], \n [0, 0, 1, 0], \n [0, 0, 0, 1]], dtype=np.float32)\n self.measurement_matrix = np.array([[1, 0, 0, 0], \n [0, 1, 0, 0]], dtype=np.float32)\n\n def predict(self):\n self.state = np.dot(self.transition_matrix, self.state)\n self.error_covariance = np.dot(np.dot(self.transition_matrix, self.error_covariance), \n self.transition_matrix.T) + self.process_noise\n return self.state[:2]\n\n def correct(self, measurement):\n kalman_gain = np.dot(np.dot(self.error_covariance, self.measurement_matrix.T), \n np.linalg.inv(np.dot(np.dot(self.measurement_matrix, self.error_covariance), \n self.measurement_matrix.T) + self.measurement_noise))\n self.state = self.state + np.dot(kalman_gain, (measurement - np.dot(self.measurement_matrix, self.state)))\n self.error_covariance = self.error_covariance - np.dot(np.dot(kalman_gain, self.measurement_matrix), self.error_covariance)\n return self. State[:2]\n\n# Perform object detection on the current frame to obtain the position of a person, such as using object detection method. Assuming we have detected the position of a person, we can update the Kalman filter to obtain the predicted and corrected position\n```\n\n\n## 2. Using some image classification/detection model Backbone, and implement your own model.\nYolov8: https://github.com/ultralytics/ultralytics\nUnet: https://github.com/topics/unet\netc.....\n\n\n## 3. You may need a POWERFUL GPU for traing.", "2259514": "is there any notebook available where the kalman filter is implemented on images as such. Link please if found. Upvoted , great discussion thread.", "2259125": "Thank you! This is very useful information. ", "2254488": "Informative, thanks for share.", "2257061": "Thanks! Could you please explain Kalman Filtering a bit more or provide resources for it?", "2327293": "", "2273859": "Thanks for sharing!", "2265126": "Thanks! Good suggestions.", "2254510": "Thanks a lot!" } }