![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4495635%2Fce506dcac3b17bde483aa714776cd071%2Foutput_figure.png?generation=1713620945846323&alt=media\")
Hello Everyone,
\nOverlap detection notebook is here
\nSimilar to the approach employed in previous years, estimating the overlap region is an adopted technique for image alignment. The method involves a two-stage matching process. In the first stage, an initial matching is preformed to identify the approximate overlap region between the two images. This region serves as a constraint for the second stage, where we conduct a more refined matching within the overlap area.
\nReference:
\nhttps://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/417407
Hello Everyone,
\nOverlap detection notebook is here
\nSimilar to the approach employed in previous years, estimating the overlap region is an adopted technique for image alignment. The method involves a two-stage matching process. In the first stage, an initial matching is preformed to identify the approximate overlap region between the two images. This region serves as a constraint for the second stage, where we conduct a more refined matching within the overlap area.
\nReference:
\nhttps://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/417407
Nice sample code. Did you also experiment with other methods to determine the overlap?
\nI recall reading in other competitions that also clustering techniques where used to determine areas of interest and overlap.
\nPersonally I think this overlap method is more simple and practical.
", "rawMarkdown": "Nice sample code. Did you also experiment with other methods to determine the overlap?\n\nI recall reading in other competitions that also clustering techniques where used to determine areas of interest and overlap.\n\nPersonally I think this overlap method is more simple and practical.", "votes": null }, { "id": "2766259", "postDate": "04/21/2024 15:19:25", "content": "Thank you.
\nI haven't experimented yet with other overlap techniques, not sure if I will, so much ground to cover.
It's so grateful for such nice sample code you posted, but I wonder how can I use it in the exacte compitation?
\nI can find a proper way to combine it to the baseline code
You are very welcome!
\nYou would have to concatenate keypoints and matches all togother.
():\n \n keypoints_merged, matches_merged\n😁Let me try! thank you so much!!!
\nBut It seems that a single image will contains in many image pairs, concatenate them will got a extremely huge ndarry😱
Many keypoints will be repetitive, so you can remove duplicates and keep unique matches.
", "rawMarkdown": "Many keypoints will be repetitive, so you can remove duplicates and keep unique matches.", "votes": null } ], "comments": [ { "id": 2766099, "author_name": "rsmits", "author_url": "", "post_date": "04/21/2024 13:49:07", "content": "Nice sample code. Did you also experiment with other methods to determine the overlap?
\nI recall reading in other competitions that also clustering techniques where used to determine areas of interest and overlap.
\nPersonally I think this overlap method is more simple and practical.
", "votes": null, "replies": [ { "id": 2766259, "author_name": "nartaa", "author_url": "", "post_date": "04/21/2024 15:19:25", "content": "Thank you.
\nI haven't experimented yet with other overlap techniques, not sure if I will, so much ground to cover.
It's so grateful for such nice sample code you posted, but I wonder how can I use it in the exacte compitation?
\nI can find a proper way to combine it to the baseline code
You are very welcome!
\nYou would have to concatenate keypoints and matches all togother.
():\n \n keypoints_merged, matches_merged\n😁Let me try! thank you so much!!!
\nBut It seems that a single image will contains in many image pairs, concatenate them will got a extremely huge ndarry😱
Many keypoints will be repetitive, so you can remove duplicates and keep unique matches.
", "votes": null, "replies": [] } ] } ] } ] } ], "raw_markdown_by_id": { "2763356": "Hello Everyone,\n\nOverlap detection notebook is [here](https://www.kaggle.com/nartaa/imc24-overlap-detection)\n\nSimilar to the approach employed in previous years, estimating the overlap region is an adopted technique for image alignment. The method involves a two-stage matching process. In the first stage, an initial matching is preformed to identify the approximate overlap region between the two images. This region serves as a constraint for the second stage, where we conduct a more refined matching within the overlap area.\n\nReference:\nhttps://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/417407\n\n", "2766099": "Nice sample code. Did you also experiment with other methods to determine the overlap?\n\nI recall reading in other competitions that also clustering techniques where used to determine areas of interest and overlap.\n\nPersonally I think this overlap method is more simple and practical.", "2766259": "Thank you. \nI haven't experimented yet with other overlap techniques, not sure if I will, so much ground to cover.", "2800718": "It's so grateful for such nice sample code you posted, but I wonder how can I use it in the exacte compitation?\nI can find a proper way to combine it to the baseline code", "2801173": "You are very welcome!\nYou would have to concatenate keypoints and matches all togother.\n\n```python\ndef merge_keypoints_matches(keypoints, matches, keypoints_new, matches_new):\n # merge all keypoints and matches ... you could use np.concatenate\n return keypoints_merged, matches_merged\n```", "2801257": "😁Let me try! thank you so much!!!\nBut It seems that a single image will contains in many image pairs, concatenate them will got a extremely huge ndarry😱", "2801434": "Many keypoints will be repetitive, so you can remove duplicates and keep unique matches." }, "source": "meta" }