huahuabai
huahuabai
您好，请问您在训练CTW数据集时有没有遇到这个问题呀 
> Can you please guide me on how you convert the MJSynth to lmdb given their current annotation format? 请问您解决了吗
> 在dataset/pan_ctw.py的prepare_test_data函数中有关于img_meta的字段写入，只需要添加 > > ``` > def prepare_test_data(self, index): > img_path = self.img_paths[index] > # print('img_path:', img_path) > img_name = img_path.split('/')[-1]#修改的地方 > img = get_img(img_path, self.read_type) > img_meta = dict(org_img_size=np.array(img.shape[:2]))...
> > I train the model on ctw1500,MSRA-TD500,Total-text but i'm trying to do visualization can anyone help me with this please > > Hello, have you succeeded? I also need...
请问您找到了吗
> 你们epoch都设置24么？早停机制我觉得没啥问题吧，按照作者的训练方法，应该就是把最后的那个h5再继续训练 请问第一次训练的时候，运行preprocess和label之后，在cfg里面选择不加载模型，等训练结束选择valloss最小的作为384的预训练模型并重命名为east_model_weights_3T384.h5,每次都这样重复，请问我这样的训练方法是正确的吗
Hello, do you know how to train now, I have the same question as you, can you guide me? Thank you!
请问您现在知道该如何进行训练了吗
Hello, do you know how to train now, I have the same question as you, can you guide me?