draw.py

'''
BSD 3-Clause License

Copyright (c) 2020, Lawrence Livermore National Laboratory
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice, this
   list of conditions and the following disclaimer.

2. Redistributions in binary form must reproduce the above copyright notice,
   this list of conditions and the following disclaimer in the documentation
   and/or other materials provided with the distribution.

3. Neither the name of the copyright holder nor the names of its
   contributors may be used to endorse or promote products derived from
   this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
'''
'''
https://github.com/LLNL/fastcam

A toolkit for efficent computation of saliency maps for explainable 
AI attribution.

This work was performed under the auspices of the U.S. Department of Energy 
by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344 
and was supported by the LLNL-LDRD Program under Project 18-ERD-021 and 
Project 17-SI-003. 

Software released as LLNL-CODE-802426.

See also: https://arxiv.org/abs/1911.11293
'''
import cv2
import numpy as np
import math
import torch

try:
    from . import misc
except ImportError:
    import misc

# *******************************************************************************************************************
class _Draw(object):
    r'''
        This is a base class for drawing different representations of saliency maps.
        
        See the derived classes below for descriptions of parameters.
    '''
    
    def __init__(self, shape, weights, color):
        
        # torch tensor will be callable
        if callable(shape):
            shape = [shape()[1], shape()[2], shape()[0]]
        
        assert isinstance(shape,tuple) or isinstance(shape,list), "Output shape should be a list or tuple"
        assert len(shape) == 3, "Output shape should be height x width x chan"
        assert isinstance(color,int), "Color should be an OpenCV defined integer"   # How openCV treats this        
        
        self.height  = shape[0]
        self.width   = shape[1]
        self.chans   = shape[2]
        
        self.color   = color 
                
        assert self.height   > 0, "Height should be greater than 0"
        assert self.width    > 0, "Width should be greater than 0"
        assert self.chans    > 0, "Chans should be greater than 0"
        
        if weights is None:
            self.weights = np.array([1.0 for _ in range(self.chans)]).astype(np.float32)
        elif len(weights) == 1:
            assert weights[0] > 0, "is specified, single weight should be greater than 0"
            self.weights = np.array([weights[0] for _ in range(self.chans)]).astype(np.float32)  
        else:
            assert len(weights) == self.chans, "List of weights should be the same size as output channels"
            self.weights = np.array(weights).astype(np.float32)
                
        self.fc             = float(self.chans)
        self.frac           = 1.0/self.fc
        self.HSV_img        = np.empty((self.height,self.width,3), dtype=np.float32)
        
        self.sum_weights    = np.sum(self.weights)
        
    def _range_normalize(self, data):
        
        norm = data.max() - data.min()
        if norm != 0:
            data = (data - data.min()) / norm
            
        return data
    
    def __call__(self, input_patches):
        
        return self.make(input_patches)
    
# *******************************************************************************************************************        
class HeatMap(_Draw):        
    r'''
        This will create a heat map from a stack of saliency maps. in a H x W x C numpy array. 
        
        Parameters
       
           shape:    This is a list (H,W,C) of the expected size of the saliency map.
           weights:  This is a list of length C of weight for each channel.
           color:    The color conversion method to use. Default: cv2.COLOR_HSV2BGR 
           
        Returns:
        
            An openCV compatible numpy array sized H x W x 3 
    ''' 
        
    def __init__(self, shape, weights=None, color=cv2.COLOR_HSV2BGR):
        
        super(HeatMap, self).__init__(shape=shape, weights=weights, color=color)
        
        self.Cexp        = 1.0/math.exp(1.0)

    def make(self, input_patches, blend_img=None):
        r'''
            Input:
                input_patches: A numpy array. It should be sized [height x width x channels]. Here channels is each saliency map.
            Returns:
                A numpy array sized [height x width x 3].
        '''
        if torch.is_tensor(input_patches):
            input_patches = misc.TensorToNumpyImages(input_patches)
        if torch.is_tensor(blend_img):
            blend_img = misc.TensorToNumpyImages(blend_img)
        
        assert blend_img is None or isinstance(blend_img, np.ndarray), "Blend Image should be a numpy array or torch tensor"    
        assert isinstance(input_patches, np.ndarray), "Input should be a numpy array"
        assert len(input_patches.shape) == 3, "Input should be height x width x chan"
        assert input_patches.shape[0] == self.height, "Input should be height x width x chan"
        assert input_patches.shape[1] == self.width, "Input should be height x width x chan"
        assert input_patches.shape[2] == self.chans, "Input should be height x width x chan"
        
        patches             = self._range_normalize(input_patches.astype(np.float32)) * self.weights
        
        r'''
            Set intensity to be the weighted average
        '''
        V                   = np.sum(patches, 2) / self.sum_weights
        V                   /= V.max()    
        
        r'''
            Use the standard integral method for saturation, but give it a boost.
        '''
        if self.frac != 1.0:
            S                   = 1.0 - (np.sum(patches,2)/(self.fc*np.amax(patches,2)) - self.frac)*(1.0/(1.0 - self.frac))
        else:
            S                   = V
            
        S                   = pow(S,self.Cexp)
        
        r'''
            Set H,S and V in that order. 
        '''
        self.HSV_img[:,:,0] = (1.0 - V) * 240.0 
        self.HSV_img[:,:,1] = S
        self.HSV_img[:,:,2] = V
        
        img = cv2.cvtColor(self.HSV_img,  self.color)

        if blend_img is not None:
            img = cv2.resize(img, (blend_img.shape[0],blend_img.shape[1]))
            return misc.AlphaBlend(img, blend_img)
        else:
            return img

# *******************************************************************************************************************
class LOVI(_Draw):        
    r'''
        This will create a LOVI map from a stack of saliency maps. in a H x W x C numpy array. 
        
        Parameters
       
           shape:    This is a list (H,W,C) of the expected size of the saliency map.
           weights:  This is a list of length C of weight for each channel.
           color:    The color conversion method to use. Default: cv2.COLOR_HSV2BGR 
           
        Returns:
        
            An openCV compatible numpy array sized H x W x 3 
    ''' 
    def __init__(self, shape, weights=None, color=cv2.COLOR_HSV2BGR):
        
        super(LOVI, self).__init__(shape=shape, weights=weights, color=color)
        
        self.pos_img    = np.empty((self.height,self.width,self.chans),     dtype=np.float32)
        
        y = 1.0/((self.fc - 1.0)/self.fc)
        
        for c_i in range(self.chans): 
            self.pos_img[:,:,c_i]   = 1.0 - (float(c_i)/(self.fc))*y    
        
    def make(self, input_patches, blend_img=None):
        r'''
            Input:
                input_patches: A numpy array. It should be sized [height x width x channels]. Here channels is each saliency map.
            Returns:
                A numpy array sized [height x width x 3].
        '''
        if torch.is_tensor(input_patches):
            input_patches = misc.TensorToNumpyImages(input_patches)
        if torch.is_tensor(blend_img):
            blend_img = misc.TensorToNumpyImages(blend_img)
       
        assert blend_img is None or isinstance(blend_img, np.ndarray), "Blend Image should be a numpy array or torch tensor"             
        assert isinstance(input_patches, np.ndarray), "Input should be a numpy array or torch tensor"
        assert len(input_patches.shape) == 3, "Input should be height x width x chan"
        assert input_patches.shape[0] == self.height, "Input should be height x width x chan"
        assert input_patches.shape[1] == self.width, "Input should be height x width x chan"
        assert input_patches.shape[2] == self.chans, "Input should be height x width x chan"
        
        patches             = self._range_normalize(input_patches.astype(np.float32)) * self.weights
        
        r'''
            Compute position
        '''
        pos                 = patches * self.pos_img
        
        r'''
            Get Mean
        '''
        m                   = np.sum(pos,2) / np.sum(patches,2)

        r'''
            Set H,S and V in that order.   
        '''     
        self.HSV_img[:,:,0] = m*300
        self.HSV_img[:,:,1] = 1.0 - (np.sum(patches,2)/(self.fc*np.amax(patches,2)) - self.frac)*(1.0/(1.0 - self.frac))
        self.HSV_img[:,:,2] = np.amax(patches, 2)
          
        img = cv2.cvtColor(self.HSV_img,  self.color)
        
        if blend_img is not None:
            img = cv2.resize(img, (blend_img.shape[0],blend_img.shape[1]))
            return misc.AlphaBlend(img, blend_img)
        else:
            return img