2h4dl

@rainLiuplus https://drive.google.com/file/d/1zizj49H0pdkWmEXshPaXI3qtUd7r3cbG/view?usp=sharing, plz check it.

Hi @JiahaoYao. You mean tf.layers.batch_normalization as same as tf.contrib.layers.batch_norm? After resave model, some variables lost, model is smaller than before. How to use batch norm in tensorflow to avoid it?

Hi @JiahaoYao. As you mentioned, after resave the model, `tf.cond` is gone. But still have a question, sorry about that. In this [wiki](https://github.com/Microsoft/MMdnn/wiki/Tensorflow-parser-problem-(Slim-and-tf.layers)), it says `tf.cond` exists in `slim` not...

> ``` > if abs(video_clip.rotation) in (90, 270): > video_clip = video_clip.resize(answer_clip.size[::-1]) > video_clip.rotation = 0 > ``` Thx @chenbin0522, it works.

@jcupitt Thank you for you help, this helps me a lot.

Binary data I guess, cause need to uplaod an image over the network, not just PNG, also JPEG, GIF and other formats.

I would like to try the raw binary data , how to get it? Trying to read C++ API, but hard to me.

`write_to_memory` gives a `` , but nothing can do without read it. If I convert it to a `numpy.ndarry`, it will lose image attribute info, like type, format, exif and...

 I need to detect colorspace of this image. If its colorspace is not sRGB, I will convert it to sRGB with `pyvips.Image.icc_transform`, else do nothing.

Thanks your answer @jcupitt. Still have a question. I'm trying to read image from internet without knowing its real format, maybe gif image but named a jpg, how to read...