{ "id": 294221, "title": "Articles about Small Object Detection", "url": "/competitions/tensorflow-great-barrier-reef/discussion/294221", "author_name": "Bilzard", "post_date": "2021-12-09T08:12:23.153000", "votes": 5, "comment_count": 0, "views": 0, "content": "

The below picture (Fig. 1) shows the bounding boxes proposed by a YOLOv5 model trained by this competitions dataset.
\nSeeing this picture, we can find that confidence value of tiny object prediction is extremely small.
\nThat means tiny object is hard to detect.

\n

Fig1. proposed bounding boxes by YOLOv5 model trained by this competitions dataset
\n\"Screen-Shot-2021-12-09-at-12-39-44\"

\n

We should investigate further whether the low confidence of small objects really matters, but it is important to know what we can do to solve the problem.

\n

So I picked up some articles about small object detection.

\n\n

Besides, though they are not scientific papers, the blog listed below are also helped (all of the article listed above is cited here).

\n", "messages": [ { "id": 1612682, "postDate": "2021-12-09T08:12:23.153Z", "content": "

The below picture (Fig. 1) shows the bounding boxes proposed by a YOLOv5 model trained by this competitions dataset.
\nSeeing this picture, we can find that confidence value of tiny object prediction is extremely small.
\nThat means tiny object is hard to detect.

\n

Fig1. proposed bounding boxes by YOLOv5 model trained by this competitions dataset
\n\"Screen-Shot-2021-12-09-at-12-39-44\"

\n

We should investigate further whether the low confidence of small objects really matters, but it is important to know what we can do to solve the problem.

\n

So I picked up some articles about small object detection.

\n\n

Besides, though they are not scientific papers, the blog listed below are also helped (all of the article listed above is cited here).

\n", "rawMarkdown": "The below picture (Fig. 1) shows the bounding boxes proposed by a YOLOv5 model trained by this competitions dataset.\nSeeing this picture, we can find that confidence value of tiny object prediction is extremely small.\nThat means tiny object is hard to detect.\n\nFig1. proposed bounding boxes by YOLOv5 model trained by this competitions dataset\n\"Screen-Shot-2021-12-09-at-12-39-44\"\n\nWe should investigate further whether the low confidence of small objects really matters, but it is important to know what we can do to solve the problem.\n\nSo I picked up some articles about small object detection.\n\n* [Finding Tiny Faces](https://arxiv.org/pdf/1612.04402.pdf)\n* [Small Object Detection in Optical Remote Sensing Images via Modified Faster RCNN\n](https://www.researchgate.net/publication/325268539_Small_Object_Detection_in_Optical_Remote_Sensing_Images_via_Modified_Faster_R-CNN)\n* [Small Object Detection with Multiscale Features](https://www.hindawi.com/journals/ijdmb/2018/4546896/)\n* [YOLO-Fine: One-Stage Detector of Small Objects Under Various Backgrounds in Remote Sensing Images](https://www.mdpi.com/2072-4292/12/15/2501)\n* [HRDNet: High-resolution Detection Network for Small Objects](https://arxiv.org/abs/2006.07607)\n* [Small Object Detection with Multiscale Features](https://www.hindawi.com/journals/ijdmb/2018/4546896/)\n\nBesides, though they are not scientific papers, the blog listed below are also helped (all of the article listed above is cited here).\n\n* https://blog.roboflow.com/detect-small-objects/\n* https://binginagesh.medium.com/small-object-detection-an-image-tiling-based-approach-bce572d890ca#23fd\n* https://medium.datadriveninvestor.com/small-objects-detection-problem-c5b430996162", "votes": 5 } ], "comments": [], "raw_markdown_by_id": { "1612682": "The below picture (Fig. 1) shows the bounding boxes proposed by a YOLOv5 model trained by this competitions dataset.\nSeeing this picture, we can find that confidence value of tiny object prediction is extremely small.\nThat means tiny object is hard to detect.\n\nFig1. proposed bounding boxes by YOLOv5 model trained by this competitions dataset\n\"Screen-Shot-2021-12-09-at-12-39-44\"\n\nWe should investigate further whether the low confidence of small objects really matters, but it is important to know what we can do to solve the problem.\n\nSo I picked up some articles about small object detection.\n\n* [Finding Tiny Faces](https://arxiv.org/pdf/1612.04402.pdf)\n* [Small Object Detection in Optical Remote Sensing Images via Modified Faster RCNN\n](https://www.researchgate.net/publication/325268539_Small_Object_Detection_in_Optical_Remote_Sensing_Images_via_Modified_Faster_R-CNN)\n* [Small Object Detection with Multiscale Features](https://www.hindawi.com/journals/ijdmb/2018/4546896/)\n* [YOLO-Fine: One-Stage Detector of Small Objects Under Various Backgrounds in Remote Sensing Images](https://www.mdpi.com/2072-4292/12/15/2501)\n* [HRDNet: High-resolution Detection Network for Small Objects](https://arxiv.org/abs/2006.07607)\n* [Small Object Detection with Multiscale Features](https://www.hindawi.com/journals/ijdmb/2018/4546896/)\n\nBesides, though they are not scientific papers, the blog listed below are also helped (all of the article listed above is cited here).\n\n* https://blog.roboflow.com/detect-small-objects/\n* https://binginagesh.medium.com/small-object-detection-an-image-tiling-based-approach-bce572d890ca#23fd\n* https://medium.datadriveninvestor.com/small-objects-detection-problem-c5b430996162" } }