Set config keys via list (e.g., from command line).

Load category, model names from a shapenet dataset.

Compute bounding box of vertex array.

Translate array of vertices by vector t.

Creates and returns an instance of the model given its class name. The created model has a single placeholder node for feeding images.

Make a set of data processes for parallel data loading.

Render the object

Generate the reconstruction, loss, and activation. Evaluate loss if ground truth output is given. Otherwise, return reconstruction and

Parse MTL file and create mapping between its materials and OBJ materials. Eventual edge cases are handled here (missing materials, missing MTL

Generate hex color corresponding to integer. Colors should have well defined ordering. First N colors are hardcoded, then colors are random

Create model name based of filename ("path/fname.js" -> "fname").

Takes numpy.array img

y must be a tensor that has the same dimensions as the input. For each channel, only one element is one indicating the ground truth p

Parse OBJ file.

Read binvox header. Mostly meant for internal use.

transparency is either 'SKY', 'TRANSPARENT' If set 'SKY', render background using sky color.

Main training function

ADAM update

Stochastic Gradient Descent w/ momentum

data_queue : Multiprocessing queue data_paths : list of data and label pair used to load data repeat : if set True, return da

Merge config dictionary a into config dictionary b, clobbering the options in b whenever they are also specified in a.

Align bottom of the model with the floor (Y-axis) and center it around X and Z.

Center model (middle of bounding box).

Center model around X and Z.

Load a config file and merge it into the default options.

Extract colors from materials and assign them to faces

Extract diffuse colors from MTL materials

Return true if file exists and is accessible for reading. Should be safer than just testing for existence due to links and permissions magic o

Generate JS array of materials objects JS material objects are basically prettified one-to-one mappings of MTL properties in JSON format.

Generate dummy materials (if there is no MTL file).

Returns a tuple of sample models.

Main demo function

Test function

Test function

Return model names

A child network must define self.loss self.error self.params self.output self.activations is optional

Parse MTL file.

Parse text chunk specifying single vertex. Possible formats: vertex index vertex index / texture index vertex index / text

Read binary binvox format as array. Returns the model with accompanying metadata. Voxels are stored in a three-dimensional numpy array, whi

Save the current network parameters to the save_dir and make a symlink to the latest param so that the training function can easily l

Override the function