\n
![\"results-26-2\"](\"https://i.ibb.co/hBkwtVG/results-26-2.png\")
I will share example code of Seq-NMS[1]:
\nhttps://www.kaggle.com/tatamikenn/example-code-of-seq-nms
Although prediction code is private, you can imagine how Seq-NMS works with real-time situation.
\nYou can see newly detected boxes by Seq-NMS as the yellow boxes in the below picture.
\nI shows confidence of boxes are calibrated to high value when nearby boxes exists in the past frames.
This is more robust algorithm compared to SORT, because it re-uses the model prediction. It predict more tight-fit boxes.
\nImage frame t-1:
\n
Image frame t:
\n![\"results-26-4\"](\"https://i.ibb.co/gv6gpBy/results-26-4.png\")
[1] https://arxiv.org/abs/1602.08465
", "messages": [ { "id": 1692413, "postDate": "2022-02-16T02:45:54.043Z", "content": "I will share example code of Seq-NMS[1]:
\nhttps://www.kaggle.com/tatamikenn/example-code-of-seq-nms
Although prediction code is private, you can imagine how Seq-NMS works with real-time situation.
\nYou can see newly detected boxes by Seq-NMS as the yellow boxes in the below picture.
\nI shows confidence of boxes are calibrated to high value when nearby boxes exists in the past frames.
This is more robust algorithm compared to SORT, because it re-uses the model prediction. It predict more tight-fit boxes.
\nImage frame t-1:
\n
Image frame t:
\n
[1] https://arxiv.org/abs/1602.08465
", "rawMarkdown": "I will share example code of Seq-NMS[1]:\nhttps://www.kaggle.com/tatamikenn/example-code-of-seq-nms\n\nAlthough prediction code is private, you can imagine how Seq-NMS works with real-time situation.\n\nYou can see newly detected boxes by Seq-NMS as the yellow boxes in the below picture.\nI shows confidence of boxes are calibrated to high value when nearby boxes exists in the past frames.\n\nThis is more robust algorithm compared to SORT, because it re-uses the model prediction. It predict more tight-fit boxes.\n\n**Image frame t-1:**\n\n\n**Image frame t**:\n\n\n## Reference\n\n[1] https://arxiv.org/abs/1602.08465\n",
"votes": 7
}
],
"comments": [],
"raw_markdown_by_id": {
"1692413": "I will share example code of Seq-NMS[1]:\nhttps://www.kaggle.com/tatamikenn/example-code-of-seq-nms\n\nAlthough prediction code is private, you can imagine how Seq-NMS works with real-time situation.\n\nYou can see newly detected boxes by Seq-NMS as the yellow boxes in the below picture.\nI shows confidence of boxes are calibrated to high value when nearby boxes exists in the past frames.\n\nThis is more robust algorithm compared to SORT, because it re-uses the model prediction. It predict more tight-fit boxes.\n\n**Image frame t-1:**\n\n\n**Image frame t**:\n\n\n## Reference\n\n[1] https://arxiv.org/abs/1602.08465\n"
}
}