René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"test resolution (输入尺度带来的增益)","articleBody":"Hi Feng, I'd like to first appreciate this implementation. However, as mentioned in issue #35 , the processing of inference is different from the original [Centernet](https://github.com/xingyizhou/CenterNet/blob/master/src/lib/detectors/ctdet.py). Specifically, the original algo. resize the image to (512, 512), and then the decoded results are mapped into the original sizes. But in this implementation, **the input is padded and then sent to the network**.  I think **the major performance gain is brought by the larger resolution**, and I change the inference process like [Centernet](https://github.com/xingyizhou/CenterNet/blob/master/src/lib/detectors/ctdet.py) ( **steps: resize the image to (512, 512) -\u003e forward -\u003e decode on the heatmap (128x128) -\u003e resize via the original width/height**)， the inference results are listed as follows. \r\n\r\n | Model |  Inference Size (512x512) |  Reported (Pad \u003c=32) | [Centernet](https://github.com/xingyizhou/CenterNet.git) |\r\n | --- | --- | --- | --- |\r\n | Res18-DCN| 28.8 |29.8  | 28.1 | \r\n | Res50-DCN| 33.9 | 34.9| - |\r\n|  Res101-DCN| 35.4 | 36.8 | 34.6 | \r\n\r\nCompared to the official implementation,  the performance gain is about 0.8 points. \r\n\r\n\r\n\r\n","author":{"url":"https://github.com/JosephChenHub","@type":"Person","name":"JosephChenHub"},"datePublished":"2022-01-26T12:30:17.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":3},"url":"https://github.com/52/CenterNet-better/issues/52"}