EfficientDet is a new family of object detector models that is based on EfficientNet and is reportedly much efficient than other state of the art models.
\n\nModel efficiency has become increasingly important in computer vision. In this paper, we systematically study various neural network architecture design choices for object detection and propose several key optimizations to improve efficiency. First, we propose a weighted bi-directional feature pyramid network (BiFPN), which allows easy and fast multi-scale feature fusion; Second, we propose a compound scaling method that uniformly scales the resolution, depth, and width for all backbone, feature network, and box/class prediction networks at the same time. Based on these optimizations, we have developed a new family of object detectors, called EfficientDet, which consistently achieve an order-of-magnitude better efficiency than prior art across a wide spectrum of resource constraints. In particular, without bells and whistles, our EfficientDet-D7 achieves stateof-the-art 51.0 mAP on COCO dataset with 52M parameters and 326B FLOPS1 , being 4x smaller and using 9.3x fewer FLOPS yet still more accurate (+0.3% mAP) than the best previous detector.
\n\nInformation Source : @abhishek from his recent linkedin post\n*Paper link*
", "messages": [ { "id": 680973, "postDate": "2019-11-25T13:23:47.460Z", "content": "EfficientDet is a new family of object detector models that is based on EfficientNet and is reportedly much efficient than other state of the art models.
\n\nModel efficiency has become increasingly important in computer vision. In this paper, we systematically study various neural network architecture design choices for object detection and propose several key optimizations to improve efficiency. First, we propose a weighted bi-directional feature pyramid network (BiFPN), which allows easy and fast multi-scale feature fusion; Second, we propose a compound scaling method that uniformly scales the resolution, depth, and width for all backbone, feature network, and box/class prediction networks at the same time. Based on these optimizations, we have developed a new family of object detectors, called EfficientDet, which consistently achieve an order-of-magnitude better efficiency than prior art across a wide spectrum of resource constraints. In particular, without bells and whistles, our EfficientDet-D7 achieves stateof-the-art 51.0 mAP on COCO dataset with 52M parameters and 326B FLOPS1 , being 4x smaller and using 9.3x fewer FLOPS yet still more accurate (+0.3% mAP) than the best previous detector.
\n\nInformation Source : @abhishek from his recent linkedin post\n*Paper link*
", "rawMarkdown": "EfficientDet is a new family of object detector models that is based on EfficientNet and is reportedly much efficient than other state of the art models.\n\n\n\n**Model efficiency has become increasingly important in computer vision. In this paper, we systematically study various neural network architecture design choices for object detection and propose several key optimizations to improve efficiency. First, we propose a weighted bi-directional feature pyramid network (BiFPN), which allows easy and fast multi-scale feature fusion; Second, we propose a compound scaling method that uniformly scales the resolution, depth, and width for all backbone, feature network, and box/class prediction networks at the same time. Based on these optimizations, we have developed a new family of object detectors, called EfficientDet, which consistently achieve an order-of-magnitude better efficiency than prior art across a wide spectrum of resource constraints. In particular, without bells and whistles, our EfficientDet-D7 achieves stateof-the-art 51.0 mAP on COCO dataset with 52M parameters and 326B FLOPS1 , being 4x smaller and using 9.3x fewer FLOPS yet still more accurate (+0.3% mAP) than the best previous detector.**\n\nInformation Source : @abhishek from his recent linkedin post\n****[Paper link](https://github.com/mobassir94/ML-DL-Research-paper-Collection/blob/master/3D%20Deep%20Learning/EfficientDet.pdf)****\n\n", "votes": 11 }, { "id": 717091, "postDate": "2020-01-12T17:59:55.147Z", "content": "I just tried. Didn't seem to work.
", "rawMarkdown": "I just tried. Didn't seem to work.", "votes": 1 }, { "id": 689946, "postDate": "2019-12-07T17:56:33.247Z", "content": "Wow, this model is wonderful, thanks for sharing, I'll try it.
", "rawMarkdown": "Wow, this model is wonderful, thanks for sharing, I'll try it. ", "votes": 1, "replies": [ { "id": 689953, "postDate": "2019-12-07T18:20:57.603Z", "content": "thanks @diegojohnson \ni wanted to try it but i didn't find source code in github :(
", "rawMarkdown": "thanks @diegojohnson \ni wanted to try it but i didn't find source code in github :(", "votes": 1 }, { "id": 690168, "postDate": "2019-12-08T05:02:36.250Z", "content": "I find some repo: @mobassir \nhttps://www.kaggle.com/diegojohnson/best-algorithm-so-far-google-s-new-paper-in-nov
", "rawMarkdown": "I find some repo: @mobassir \nhttps://www.kaggle.com/diegojohnson/best-algorithm-so-far-google-s-new-paper-in-nov", "votes": 1 } ] }, { "id": 681427, "postDate": "2019-11-26T05:17:42.003Z", "content": "Thanks for Sharing @mobassir
", "rawMarkdown": "Thanks for Sharing @mobassir ", "votes": 1, "replies": [ { "id": 681434, "postDate": "2019-11-26T05:29:22.330Z", "content": "You are welcome
", "rawMarkdown": "You are welcome " } ] }, { "id": 711760, "postDate": "2020-01-06T14:25:31.123Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 717091, "author_name": "Gold Retriever", "author_url": "", "post_date": "2020-01-12T17:59:55.147000", "content": "I just tried. Didn't seem to work.
", "votes": 1, "replies": [] }, { "id": 689946, "author_name": "DiegoJohnson", "author_url": "", "post_date": "2019-12-07T17:56:33.247000", "content": "Wow, this model is wonderful, thanks for sharing, I'll try it.
", "votes": 1, "replies": [ { "id": 689953, "author_name": "Mobassir", "author_url": "", "post_date": "2019-12-07T18:20:57.603000", "content": "thanks @diegojohnson \ni wanted to try it but i didn't find source code in github :(
", "votes": 1, "replies": [] }, { "id": 690168, "author_name": "DiegoJohnson", "author_url": "", "post_date": "2019-12-08T05:02:36.250000", "content": "I find some repo: @mobassir \nhttps://www.kaggle.com/diegojohnson/best-algorithm-so-far-google-s-new-paper-in-nov
", "votes": 1, "replies": [] } ] }, { "id": 681427, "author_name": "Ailurophile", "author_url": "", "post_date": "2019-11-26T05:17:42.003000", "content": "Thanks for Sharing @mobassir
", "votes": 1, "replies": [ { "id": 681434, "author_name": "Mobassir", "author_url": "", "post_date": "2019-11-26T05:29:22.330000", "content": "You are welcome
", "votes": 0, "replies": [] } ] }, { "id": 711760, "author_name": "", "author_url": "", "post_date": "2020-01-06T14:25:31.123000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "680973": "EfficientDet is a new family of object detector models that is based on EfficientNet and is reportedly much efficient than other state of the art models.\n\n\n\n**Model efficiency has become increasingly important in computer vision. In this paper, we systematically study various neural network architecture design choices for object detection and propose several key optimizations to improve efficiency. First, we propose a weighted bi-directional feature pyramid network (BiFPN), which allows easy and fast multi-scale feature fusion; Second, we propose a compound scaling method that uniformly scales the resolution, depth, and width for all backbone, feature network, and box/class prediction networks at the same time. Based on these optimizations, we have developed a new family of object detectors, called EfficientDet, which consistently achieve an order-of-magnitude better efficiency than prior art across a wide spectrum of resource constraints. In particular, without bells and whistles, our EfficientDet-D7 achieves stateof-the-art 51.0 mAP on COCO dataset with 52M parameters and 326B FLOPS1 , being 4x smaller and using 9.3x fewer FLOPS yet still more accurate (+0.3% mAP) than the best previous detector.**\n\nInformation Source : @abhishek from his recent linkedin post\n****[Paper link](https://github.com/mobassir94/ML-DL-Research-paper-Collection/blob/master/3D%20Deep%20Learning/EfficientDet.pdf)****\n\n", "717091": "I just tried. Didn't seem to work.", "689946": "Wow, this model is wonderful, thanks for sharing, I'll try it. ", "681427": "Thanks for Sharing @mobassir ", "711760": "" } }