Prince Wang

This repo stop updating for more than one year. If someone still interesting in pytorch R-FCN, pls look at another pytorch implement - [princewang1994/R-FCN.pytorch](https://github.com/princewang1994/R-FCN.pytorch), thanks~

I fixed this problem by modify a few lines [here](https://github.com/xingmimfl/pytorch_RFCN/blob/3baed2001fe2ed47cb80ea4193ec945bbc162c06/test.py#L115) bug version(original): ```python # skip j = 0, because it's the background class for j in xrange(1, imdb.num_classes): inds =...

I got about 0.36 mAP by running train.py directly, it seems that R-FCN's structure in these repo is not the same as that in original paper such as the size...

@AmrutaAnalytics Did you use your own dataset? Seems that your input image are gray-scale ones, If so, make it RGB format first.

@smartcatdog I had made total text data available at my google drive so you can directly download by running [this file](https://github.com/princewang1994/TextSnake.pytorch/blob/master/dataset/total_text/download.sh). With some unknown error, you can also manually download...

@blueclowd I had ignore polygons with too few vertices (

What dataset you are training on? if total-text or icdar2013/15, the configuration in README can reproduce the result in table. If not in two datasets above, please check dataloader you...

It seems that model is underfitting but not overfitting. The loss should be smaller than 0.1. For small dataset overfit, you can try smaller learning rate and training more epochs.

@dongzhi0312 这个地方代码中有点小问题，在demo.py里面[这个部分](https://github.com/princewang1994/TextSnake.pytorch/blob/master/demo.py#L85)，这了predict和gt使用了相同的coutour变量，其实下面那张图片应该是使用meta变量中的坐标点的，你发的这张图片的上下两个应该都是predict，我会在接下来的版本中改正~

Thank you for your praise! I have not tried resnet50 yet(`network/resnet` is an incomplete version), if you are going to do this, try importing resnet into textnet.py. In the textnet.py...