sql

LayoutReader: Evaluation metric ARD

1

Hello, thanks for the impressive work. I want to ask the calculation of the evaluation metric "ARD" in the reported paper. Do you calculate the ARD by dividing the number...

Can this project support extract warping rotated ROI while also keeping original aspect ratio?

weakly supervised training?

2

Will you share the details of the weakly supervised part? Thanks.

Training Chinese?

1

Hello, @Pay20Y Do you have any plans for training Chinese textline? What Chinese pre-training model do you recommend?

字体文件下载后还是报错

2

你好，windows下字体文件的相对路径应该怎么放呢？下载了字体文件在C盘里，还是报错。 可以选择相对路径存吗？目录的结构大概是怎么样的？

直接加载训练好的模型，也不能复现性能

1

@SakuraRiven 你好，使用直接训练好的EAST vgg模型，也不能复现性能，请问是什么问题呢？ 我的LANMS模块从原作者里面的工程直接编译得到，应该没有影响吧. [https://github.com/argman/EAST/tree/master/lanms](url)

Reference code of pre-training tasks

4

Thanks for your impressive work. I have one question. will you share the code of other pre-training tasks? Thanks for your reply. @uakarsh @shabie

The relationship between contrastive accuracy and linear evaluation performance?

10

Hello, @HobbitLong I have succeeded in training with cifar-10 dataset. And now I train with these methods on the custom dataset. For example, when I apply MoCo method, I change...

Bug: rotate_iou CUDA mode

7

## 🐛 Bug ``` from maskrcnn_benchmark.layers.rotate_nms import rotate_iou a=torch.tensor([[50.7325,50.7325,100.7325,300.7325,-3.4]]).float().to('cuda') b=torch.tensor([[50.7325,50.7325,100.7325,300.7325,-3.4]]).float().to('cuda') rotate_iou(a,b) # get the result 0.3333 a=torch.tensor([[50.7325,50.7325,100.7325,300.7325,-3.4]]).float() b=torch.tensor([[50.7325,50.7325,100.7325,300.7325,-3.4]]).float() rotate_iou(a,b) # get the result 1.0 ``` however, when I test on...

LayoutReader: How can I inference if I only have layout information data?

22

Hello, thanks for the impressive work. When I do decode_seq2seq, it seems I need to prepare the json file with text information. I want to ask how I can do...