funding

.github/FUNDING.yml

@@ -0,0 +1,2 @@
+patreon: GreenWizard
+buy_me_a_coffee: greenwizard89

FUNDING.md

@@ -0,0 +1,9 @@
+# Support and Funding
+
+If you want to support my quest for sarcastic humor, brilliant software achievements, and witty commentary, here’s how you can do it:
+
+1. **Patreon**: All my amazing content there is free. But if you feel an irresistible urge to support me with monthly donations, head over to [my Patreon page](https://www.patreon.com/GreenWizard). Your support will help me keep delighting you with sarcasm and software revelations.
+
+2. **Buy Me a Coffee**: Prefer one-time acts of generosity? You can buy me a coffee at [Buy Me a Coffee](https://buymeacoffee.com/greenwizard89). Because, honestly, my level of sarcasm and ability to write genius code directly correlate with the amount of caffeine I consume.
+
+By supporting me, you ensure a continuous flow of sarcasm, wit, and cutting-edge software insights. And who wouldn't want more of that in their life?

NN/restorators/CARProcess.py

@@ -16,14 +16,16 @@ def __init__(self, predictions, sourceDistribution, sampler):
   def forward(self, x0, xT=None):
     B = tf.shape(x0)[0]
     # source distribution need to know the shape of the input, so we need to ensure it explicitly
-    x0 = tf.ensure_shape(x0, (None, self.predictions))
+    # x0 = tf.ensure_shape(x0, (None, self.predictions))
     sampled = self._sourceDistribution.sampleFor(x0)
     x1 = sampled['xT'] if xT is None else xT
 
-    tf.assert_equal(tf.shape(x0), (B, self.predictions))
+    # tf.assert_equal(tf.shape(x0), (B, self.predictions))
     return self._sampler.train(x0=x0, x1=x1, T=sampled['T'], xT=xT)
 
   def calculate_loss(self, x_hat, predicted, **kwargs):
+    if hasattr(self._sampler, 'calculate_loss'):
+      return self._sampler.calculate_loss(x_hat, predicted, **kwargs)
     lossFn = kwargs.get('lossFn', tf.losses.mae) # default loss function
     target = self.withExtraOutputs(x_hat['target'], **kwargs)
     tf.assert_equal(tf.shape(target), tf.shape(predicted))
@@ -45,11 +47,25 @@ def reverse(self, value, denoiser, modelT=None, index=0, **kwargs):
 
     denoiser = self._makeDenoiser(denoiser, modelT)
     res = self._sampler.sample(value=value, model=denoiser, index=index, **kwargs)
-    tf.assert_equal(tf.shape(res), tf.shape(value))
+    # tf.assert_equal(tf.shape(res), tf.shape(value))
     return res
 
   def targets(self, x_hat, values):
     return self._sampler.targets(x_hat, values[:, :self.predictions])
+
+  @property
+  def channels(self):
+    sampler = self._sampler
+    if hasattr(sampler, 'channels'):
+      return sampler.channels
+    return super().channels
+
+  @property
+  def predictions(self):
+    sampler = self._sampler
+    if hasattr(sampler, 'predictions'):
+      return sampler.predictions
+    return super().predictions
 # End of CARProcess
 
 def autoregressive_restoration_from_config(config):

NN/restorators/IRestorationProcess.py

@@ -1,6 +1,6 @@
 import tensorflow as tf
 
-class IRestorationProcess:
+class IRestorationProcess(tf.keras.Model):
   @staticmethod
   def getOutputSize(outputs):
     sizes = {
@@ -15,7 +15,9 @@ def getOutputSize(outputs):
       continue
     return channels
 
-  def __init__(self, predictions):
+  def __init__(self, predictions, name=None, **kwargs):
+    if name is None: name = self.__class__.__name__
+    super().__init__(name=name, **kwargs)
     predictions = list(predictions)
     if 'rgb' not in predictions: predictions.insert(0, 'rgb')
     self._outputs = predictions
@@ -24,6 +26,9 @@ def __init__(self, predictions):
     print('[IRestorationProcess] Restorator:', self.__class__.__name__)
     return
 
+  def call(self, *args, **kwargs):
+    raise RuntimeError('IRestorationProcess object cannot be called directly')
+
   def forward(self, x0, xT=None):
     raise NotImplementedError()
 

NN/restorators/samplers/CARSampler.py

@@ -176,7 +176,7 @@ def solve(self, x_hat, step, value, interpolant, params, **kwargs):
     )
 
   def directSolve(self, x_hat, values, interpolant, **kwargs):
-    solved = interpolant.solve(x_hat=x_hat['x1'], xt=values, t=1.0).x0
+    solved = interpolant.solve(x_hat=values, xt=x_hat['x1'], t=1.0).x0
     return solved
 # End of CARSamplingAlgorithm
 

NN/restorators/samplers/CBasicInterpolantSampler.py

@@ -2,8 +2,21 @@
 from .ISamplingAlgorithm import ISamplingAlgorithm
 from NN.utils import is_namedtuple
 
-class CBasicInterpolantSampler:
-  def __init__(self, interpolant, algorithm):
+class IBasicInterpolantSampler(tf.keras.Model):
+  @property
+  def interpolant(self):
+    raise NotImplementedError()
+
+  @tf.function
+  def sample(self, value, model, index=0, **kwargs):
+    raise NotImplementedError()
+
+  def targets(self, x_hat, values):
+    raise NotImplementedError()
+
+class CBasicInterpolantSampler(IBasicInterpolantSampler):
+  def __init__(self, interpolant, algorithm, **kwargs):
+    super().__init__(**kwargs)
     self._interpolant = interpolant
     self._algorithm = algorithm
     return
@@ -17,7 +30,7 @@ def sample(self, value, model, index=0, **kwargs):
     kwargs = dict(**kwargs, interpolant=self._interpolant, index=index)
     # wrap algorithm with hook, if provided
     algorithm = kwargs.get('algorithmInterceptor', lambda x: x)( self._algorithm )
-    assert isinstance(algorithm, ISamplingAlgorithm), f'Algorithm must be an instance of ISamplingAlgorithm, but got {type(algorithm)}'
+    assert issubclass(type(algorithm), ISamplingAlgorithm), f'Algorithm must be an instance of ISamplingAlgorithm, but got {type(algorithm)}'
 
     step = algorithm.firstStep(value=value, **kwargs)
     # CFakeObject is a namedtuple, so we need to check for it
@@ -39,6 +52,11 @@ def sample(self, value, model, index=0, **kwargs):
       # update value
       tf.assert_equal(tf.shape(value), tf.shape(solution.value))
       value = solution.value
+      # for debugging, print euclidean distance to GT
+      if 'GT' in kwargs:
+        gt = kwargs['GT']
+        dist = tf.reduce_sum(tf.square(value - gt), axis=-1)
+        tf.print(f'Iteration {iteration}:', dist, summarize=10)
       iteration += 1
       continue
 

NN/restorators/samplers/CDDIMInterpolantSampler.py

@@ -1,109 +1,6 @@
 import tensorflow as tf
-from Utils.utils import CFakeObject
-from NN.utils import normVec
-from .CBasicInterpolantSampler import CBasicInterpolantSampler, ISamplingAlgorithm
-
-class CDDIMSamplingAlgorithm(ISamplingAlgorithm):
-  def __init__(self, stochasticity, noiseProvider, schedule, steps, clipping, projectNoise):
-    self._stochasticity = stochasticity
-    self._noiseProvider = noiseProvider
-    self._schedule = schedule
-    self._steps = steps
-    self._clipping = clipping
-    self._projectNoise = projectNoise
-    return
-
-  def _makeStep(self, current_step, steps, **kwargs):
-    schedule = kwargs.get('schedule', self._schedule)
-    eta = kwargs.get('eta', self._stochasticity)
-
-    T = steps[0][current_step]
-    alpha_hat_t = schedule.parametersForT(T).alphaHat
-    prevStepInd = steps[1][current_step]
-    alpha_hat_t_prev = schedule.parametersForT(prevStepInd).alphaHat
-
-    stepVariance = schedule.varianceBetween(alpha_hat_t, alpha_hat_t_prev)
-    sigma = tf.sqrt(stepVariance) * eta
-
-    return CFakeObject(
-      steps=steps,
-      current_step=current_step,
-      active=(0 <= current_step),
-      sigma=sigma,
-      #
-      T=T,
-      t=alpha_hat_t,
-      t_prev=alpha_hat_t_prev,
-      t_prev_2=1.0 - alpha_hat_t_prev - tf.square(sigma),
-    )
-
-  def firstStep(self, **kwargs):
-    schedule = kwargs.get('schedule', self._schedule)
-    assert schedule is not None, 'schedule is None'
-    assert schedule.is_discrete, 'schedule is not discrete'
-    steps = schedule.steps_sequence(
-      startStep=kwargs.get('startStep', None),
-      endStep=kwargs.get('endStep', None),
-      config=kwargs.get('stepsConfig', self._steps),
-      reverse=True, # reverse steps order to make it easier to iterate over them
-    )
-
-    return self._makeStep(
-      current_step=tf.size(steps[0]) - 1,
-      steps=steps,
-      **kwargs
-    )
-
-  def nextStep(self, step, **kwargs):
-    return self._makeStep(
-      current_step=step.current_step - 1,
-      steps=step.steps,
-      **kwargs
-    )
-
-  def inference(self, model, step, value, **kwargs):
-    schedule = kwargs.get('schedule', self._schedule)
-    return model(
-      x=value,
-      T=step.T,
-      t=schedule.to_continuous(step.T),
-    )
-
-  def _withNoise(self, value, sigma, x0, kwargs):
-    noise_provider = kwargs.get('noiseProvider', self._noiseProvider)
-    noise = noise_provider(shape=tf.shape(value), sigma=sigma)
-    if not kwargs.get('projectNoise', self._projectNoise): return value + noise
-
-    _, L = normVec(value - x0)
-    vec, _ = normVec(value + noise - x0)
-    return x0 + L * vec # project noise back to the spherical manifold
-
-  def _withClipping(self, value, kwargs):
-    clipping = kwargs.get('clipping', self._clipping)
-    if clipping is None: return value
-    return tf.clip_by_value(value, clip_value_min=clipping['min'], clip_value_max=clipping['max'])
-
-  def solve(self, x_hat, step, value, interpolant, **kwargs):
-    solved = interpolant.solve(x_hat=x_hat, xt=value, t=step.t)
-    x_prev = interpolant.interpolate(
-      x0=solved.x0, x1=solved.x1,
-      t=step.t_prev, t2=step.t_prev_2
-    )
-    x_prev = self._withNoise(x_prev, sigma=step.sigma, x0=solved.x0, kwargs=kwargs)
-    x_prev = self._withClipping(x_prev, kwargs=kwargs)
-    # return solution and additional information for debugging
-    return CFakeObject(
-      value=x_prev,
-      x0=solved.x0,
-      x1=solved.x1,
-      T=step.T,
-      current_step=step.current_step,
-      sigma=step.sigma,
-    )
-
-  def directSolve(self, x_hat, xt, interpolant):
-    return interpolant.solve(x_hat=xt, xt=x_hat['xT'], t=x_hat['alphaHat']).x0
-# End of CDDIMSamplingAlgorithm
+from .CBasicInterpolantSampler import CBasicInterpolantSampler
+from .CDDIMSamplingAlgorithm import CDDIMSamplingAlgorithm
 
 class CDDIMInterpolantSampler(CBasicInterpolantSampler):
   def __init__(

NN/restorators/samplers/CSamplerWatcher.py

@@ -1,6 +1,5 @@
 import tensorflow as tf
 from NN.utils import is_namedtuple
-import NN.utils as NNU
 from .CSamplingInterceptor import CSamplingInterceptor
 from .ISamplerWatcher import ISamplerWatcher
 

NN/restorators/samplers/CSamplingInterceptor.py

@@ -1,24 +1,48 @@
 from .ISamplingAlgorithm import ISamplingAlgorithm
+from .ISamplerWatcher import ISamplerWatcher
 
-class CSamplingInterceptor(ISamplingAlgorithm):
+class CSamplingInterceptor(ISamplingAlgorithm, ISamplerWatcher):
   def __init__(self, watcher, algorithm):
+    assert issubclass(type(watcher), ISamplerWatcher), f'Invalid watcher: {watcher}'
     self._watcher = watcher
     self._algorithm = algorithm
     return
 
+  def interceptor(self):
+    def F(algorithm):
+      if isinstance(self._watcher, ISamplerWatcher):
+        self._watcher = algorithm = self._watcher.interceptor()(algorithm)
+
+      if callable(self._watcher): # replace the watcher with the interceptor
+        self._watcher = algorithm = self._watcher(algorithm)
+
+#      if self._algorithm is not None:
+#        assert isinstance(algorithm, ISamplingAlgorithm), f'algorithm is not an instance of ISamplingAlgorithm: {algorithm}'
+      return self
+    return F
+
   def firstStep(self, **kwargs):
-    res = self._algorithm.firstStep(**kwargs)
+    res = self._algorithm.firstStep(**kwargs) if self._algorithm is not None else None
     self._watcher._onStart(value=kwargs['value'], kwargs=kwargs)
     return res
 
+  def _onStart(self, value, kwargs):
+    return self._watcher._onStart(value=value, kwargs=kwargs)
+
   def nextStep(self, **kwargs):
     self._watcher._onNextStep(iteration=kwargs['iteration'], kwargs=kwargs)
-    res = self._algorithm.nextStep(**kwargs)
+    res = self._algorithm.nextStep(**kwargs) if self._algorithm is not None else None
     return res
 
+  def _onNextStep(self, iteration, kwargs):
+    return self._watcher._onNextStep(iteration=iteration, kwargs=kwargs)
+
   def inference(self, **kwargs):
-    return self._algorithm.inference(**kwargs)
+    return self._algorithm.inference(**kwargs) if self._algorithm is not None else None
 
   def solve(self, **kwargs):
-    return self._algorithm.solve(**kwargs)
+    return self._algorithm.solve(**kwargs) if self._algorithm is not None else None
+
+  def tracked(self, name):
+    return self._watcher.tracked(name)
 # End of CSamplingInterceptor