LI Wentong

Results 75 comments of LI Wentong

@kuazhangxiaoai `pts_coordinate_preds_init_image` 和 `pts_coordinate_preds_refine_image` 分别是intial stage 和refine stage 预测的点集在图像中实际位置的坐标。 `Q_poc` 的相关代码,请看我们现在更新的代码。

@kuazhangxiaoai `pts_preds_init` 和 `pts_preds_refine` 为`forward`函数中网络的预测,分别对应`pts_out_init`和`pts_out_refine`,代表着9个点 `x, y`上的 `offset`,故channel为`18`.

DOTA数据可以在这下载 https://captain-whu.github.io/DOTA/dataset.html

@ziruoliu Hello, the code for DOTA data preparation, please refer this code-[prepare_dota1_ms.py](https://github.com/LiWentomng/OrientedRepPoints/blob/main/DOTA_devkit/prepare_dota1_ms.py) to get single-scale image patches. The json format is similar as COCO json with 8 points(ploy) for box...

@ziruoliu Sorry to reply later. mAOE is the diferences of the detection and prediction on angular orientation . The angle definition for orientation can refer [this](https://user-images.githubusercontent.com/32033843/119216186-be2fd080-bb04-11eb-9736-1f82c6666171.png) . The codes for...

@zack2020-star 上面的提示的不匹配信息是正常的。 按照提供的swin-tiny的 config默认设置,4 GPU,可以达到78.11的结果。具体结果如下: mAP: 0.7811115629963165 ap of each class: plane:0.8869824055320612, baseball-diamond:0.8404591146489188, bridge:0.573374138252922, ground-track-field:0.7460058185240421, small-vehicle:0.8178457956880836, large-vehicle:0.8412193176830997, ship:0.8800826910103363, tennis-court:0.9076617572050741, basketball-court:0.8687596706802145, storage-tank:0.8752927349151276, soccer-ball-field:0.6482638786646695, roundabout:0.6817482393358247, harbor:0.7680666777746015, swimming-pool:0.7484314457783111, helicopter:0.6324797592514606 如果不是采用4gpu,对应的学习率应该需要调整一下。另外一个可能原因是图片裁剪方式不同,不知是否采用提供的代码进行裁剪。

你好, 这个问题解决了吗?感觉还是环境配置的问题,以及编译过程中是否正常。 目前版本的代码运行环境为 _pytorch==1.3.1,mmdet==1.1.0, mmcv==0.3.1_.

你好,训练的swin-tiny模型文件和对应的log文件已上传,可直接inference。

@Kaihui-Cheng 百度云[链接](https://pan.baidu.com/s/1J-g_OUXSyMMvJWt-xCX_0g?pwd=53zn) password: 53zn

你好,在单个2080ti上进行推理 约14FPS, RetinaNet-O的推理速度约为15.5FPS。