volume.h

#ifndef VOLUME_H
#define VOLUME_H

#include <inttypes.h>
#include <stddef.h>

// Volumes are used to represent the activations (i.e., state) between the
// different layers of the CNN. They all have three dimensions. The inter-
// pretation of their content depends on the layer that produced them. Before
// the first iteration, the Volume holds the data of the image we want to
// classify (the depth are the three color dimensions). After the last stage
// of the CNN, the Volume holds the probabilities that an image is part of
// a specific category.
//
// The weights are represented as a 1-d array with length
// width * height * depth.
typedef struct volume {
  int width;
  int height;
  int depth;
  double* weights;
} volume_t;

// Gets the element in the volume at the coordinates (x, y, d).
double volume_get(volume_t* v, int x, int y, int d);

// Sets the element in the volume at the coordinates (x, y, d) to value
void volume_set(volume_t* v, int x, int y, int d, double value);

// Allocates a new volume with the specified dimensions, initializes it to the
// specified value.
volume_t* make_volume(int width, int height, int depth, double value);

// Copies the contents of one volume into another.
void copy_volume(volume_t* dest, volume_t* src);

// Frees the weights array and the struct itself.
void free_volume(volume_t* v);

#endif