Support for more network types.

[dtmoodie]

Hello, I've been working with SSD (https://github.com/weiliu89/caffe/tree/ssd) and I would like to visualize the output with your toolkit. Currently I'd be happy with just getting a visualization of the low level activations and the fully convolutional section of the network. However I'm running into an issue in caffevis/app.py:115 because some of the layers have a 3 dimensional shape. How necessary is this assertion? What changes should be made to support or ignore 3 dimensional operations?

[chunxu-li]

@dtmoodie I just meet this problem today.

Here is what i have done for new type: I know it it not the right way to solve this problem, but for the temporary, I have no time to spend on this problem. Hope to help you!

[Bardo91]

Hi @chunxuxiao, would you mind to tell which files are those that you modified? Thank you very much!

[chunxu-li]

@Bardo91 As you can see in the line 116 of the second snapshot the assert will failed, for that the length of blob is not equal to 2 or 4 when "key =conv_4_3_norm_mbox_priorbox (As shown in the first snapshot )", I just common that 'assert' and add line 117, 118 in the py file that caused break down of the running, in another file(the third snapshot ), I add line 253 and 254 in the for loop. I'am not sure if the visualization is ritht, and sorry for forgetting the concretes name of the files, I have graduate in school and not doing that field anymore. Hope that will help you!

[arikpoz]

@dtmoodie, @chunxuxiao, @Bardo91 I've made the tool substantially more generic: all the hardcoded parts where replaced with new settings parameters, most are deduced automatically by analyzing the network definition file

Latest version is in https://github.com/arikpoz/deep-visualization-toolbox

[shonehang]

@arikpoz I want to know how to use your tool for SSD, just change the deploy prototxt and caffemodel?Thank you!!!