TextBoxes_plusplus_Tensorflow

TextBoxes_plusplus_Tensorflow copied to clipboard

pretrained model

1

Hello, The pretrained model link is dead, can you re upload it? Many thanks

stesteau

Just in case you are interested in, I made a docker on top of you code which makes it easier for other to use what you have made ...

naarkhoo

作者你好，SSD中，输入图片尺寸为300*300 ，anchor_step为[8, 16, 32, 64, 100, 300], TextBoxes++输入为384*384,是不是需要对anchor_step做一定的调整

fanw52

只能检测到一个文本框

1

我不知道是不是模型参数需要修改，我需要预测四个文本框，但是网络只能预测一个，我用训练集数据测试也只能预测一个，loss也不再下降

wy51wy

训练损失

5

 你好，这个loss一直在震荡是怎么回事啊？ loss这么高200左右对吗？

fuyunfuyun666

have trouble in textbox_anchor_one_layer

2

textbox_anchor_one_layer for generating anchors for one feature layer, but I encounter a problem while visualizing it.could some one give me some tips. thanks advance.

opconty

Hi, I just want to test my image with model without training it, so can you share your trained model such as "model.ckpt-858"? thanks

ShaneYS

例如`txtbox_384.py`的`textbox_anchor_one_layer`函数中的参数，像step这种都是应该根据输入图片的大小去计算的而不是固定设置的。 你在复现的时候是否是为了方便，而直接参考了原作者caffe的实现，所以没有考虑这方面的东西？或者说还是我的理解有问题

ofexe

测试精度问题

9

作者你好 我想问下这个tensorflow版本能达到原文精度吗？ 略次还是略高？

wangyuxin87

Could you please share your successful ssd project based on project SSD-Tensorflow of balancap. The reason is because I saw many people complaining that they can't train that model correctly...

lloydnguyen96