East to Use

python conv_vis.py            
   --model                    # Checkpoint path.
   --image_path               # Path of the input image for CNN.
   --output_dir               # Output directory for feature maps. 

The model repo should contain (for example):

export.ckpt:      Trained model parameters.
export.ckpt.meta: Model structure.

Search for Data Entrance Automatically

Given the input image, the program automatically finds all the Placeholders in the computing graph and searches for the best-matched ones to feed the image into.

Two debug trials once ThoroughVis fails:

1. Resize the input image to match the target data entrance.
2. Make sure the target Placeholder is correctly defined in the computing graph.

Default Placeholder Feed-In

The program will automatically acquire all the Placeholders and feed them with default zero values to make the computing graph flow properly.

tf.bool:         False
tf.int32:        0
tf.int64:        0
tf.float16:      0.0
tf.float32:      0.0
tf.array(shape): numpy.zeros(shape)

Our team will add self-defined feed-in support in the next update.

Minimum Requirement

tensorflow
numpy 
matplotlib
uuid

Licence

MIT