Cheni Chadowitz

Haha, awesome! It would be cool if the `best` parameter had optional configurations 🙃

That makes sense, I wasn't sure if `CaffeLogger` was original Caffe or from you :) I tried to dig through and figure out how TF's `LOG` macros changed between 0.13.1...

Interesting.... yeah, here's the relevant chunk of the dockerfile I'm using: ``` FROM ubuntu:18.04 as intermediate RUN echo 'building CPU DeepDetect image' ENV DEBIAN_FRONTEND=noninteractive COPY libopenblas-base_0.3.5+ds-2_amd64.deb libopenblas-dev_0.3.5+ds-2_amd64.deb /tmp/ RUN apt-get...

FYI I have a rough dockerfile that setups up DD on Ubuntu 18.04 (with caffe, TF, dlib backends) here: https://github.com/jolibrain/deepdetect/issues/687#issuecomment-572749679 Just make sure to use TF v0.13.1 and bazel v0.21.0...

For the `data` layer, the values matched exactly. For the next layer (`conv_1`), the values vastly diverge. I've uploaded the two outputs to gists, but they are fairly large. [Python...

Comparing the python script on CPU and GPU at the `conv_1` layer: ``` $ diff -u layer-conv_1-gpu.txt layer-conv_1-cpu.txt --- layer-conv_1-gpu.txt 2017-06-12 13:14:09.576400426 -0400 +++ layer-conv_1-cpu.txt 2017-06-12 11:30:17.987031126 -0400 @@ -363774,7...

And comparing DD CPU and DD GPU there is no numerical difference whatsoever.

I'm not sure what you mean by not see it listed in the net's layer - the one from the [official repo](https://github.com/yahoo/open_nsfw/blob/master/nsfw_model/deploy.prototxt)? It uses a standard data input layer, so...

Just an update after more lengthy investigation. There are a couple ways to do a forward pass in caffe (whether using pycaffe or the c++ api). - `Forward()`/`Forward_all()`: Given one...

Another update: It seems that with solid color images, the two approaches behave identically. When photograph-like color images are used, slight differences in pixel values when loading the image from...