def get_fid_tf(real_img, sample_img): print('debug/real_img type') print(real_img) real_img = tf.convert_to_tensor(real_img) real_img = preprocess_image(real_img) print(real_img) print('debug/sample_img type') print(sample_img) sample_img = tf.convert_to_tensor(sample_img) sample_img = preprocess_image(sample_img) print(sample_img) real_single = (real_img.shape.ndims == 3) sample_single = (sample_img.shape.ndims == 3) if real_single and sample_single: sample_img = tf.concat([sample_img, sample_img, sample_img], 3) real_img = tf.concat([real_img, real_img, real_img], 3) ll = real_img.shape[0] print('debug/ll') print(ll) pbs = 4 while ll % pbs: pbs += 1 FID = frechet_inception_distance(real_img, sample_img, num_batches=ll // pbs) return FID
def test_preprocess_image_graph(self): """Test `preprocess_image` graph construction.""" incorrectly_sized_image = array_ops.zeros([520, 240, 3]) correct_image = classifier_metrics.preprocess_image( images=incorrectly_sized_image) _run_with_mock(classifier_metrics.run_inception, array_ops.expand_dims(correct_image, 0))
def test_preprocess_image_graph(self): """Test `preprocess_image` graph construction.""" incorrectly_sized_image = array_ops.zeros([520, 240, 3]) correct_image = classifier_metrics.preprocess_image( images=incorrectly_sized_image) _run_with_mock(classifier_metrics.run_inception, array_ops.expand_dims(correct_image, 0))
def get_inception_score_tf(sample_img): print('debug/sample_img type') print(sample_img.shape) print(sample_img) sample_img = tf.convert_to_tensor(sample_img) sample_img = preprocess_image(sample_img) print(sample_img.shape) print(sample_img) ll = sample_img.shape[0] print('debug/ll') print(ll) print('debug/sample_img 2') print(sample_img.shape) pbs = 10 while True: print('debug/pbs', pbs) if ll % pbs == 0: break else: pbs -= 1 print('debug/batch_size') print(pbs) IS = classifier_score(sample_img, functools.partial( run_image_classifier, graph_def=graph_def, input_tensor=INCEPTION_INPUT, output_tensor=INCEPTION_OUTPUT, ), num_batches=ll // 100) # generated_images_list = array_ops.split( # sample_img, num_or_size_splits=ll//pbs) # # # Compute the classifier splits using the memory-efficient `map_fn`. # logits = functional_ops.map_fn( # fn=functools.partial( # run_image_classifier, # graph_def=graph_def, # input_tensor=INCEPTION_INPUT, # output_tensor=INCEPTION_OUTPUT,), # elems=array_ops.stack(generated_images_list), # parallel_iterations=1, # back_prop=False, # swap_memory=True, # name='RunClassifier') # logits = array_ops.concat(array_ops.unstack(logits), 0) # print('logits',logits) # logits.shape.assert_has_rank(2) # # # Use maximum precision for best results. # logits_dtype = logits.dtype # if logits_dtype != dtypes.float64: # logits = math_ops.to_double(logits) # # p = nn_ops.softmax(logits) # q = math_ops.reduce_mean(p, axis=0) # kl = _kl_divergence(p, logits, q) # print('pqkl',p,q,kl) # kl.shape.assert_has_rank(1) # # sess = tf.Session() # sess.run(tf.Print(kl, [kl])) # # log_score = math_ops.reduce_mean(kl) # print('log_score',log_score) # final_score = math_ops.exp(log_score) # print('final',final_score) # if logits_dtype != dtypes.float64: # final_score = math_ops.cast(final_score, logits_dtype) # IS = final_score print('debug/end IS') IS = IS.shape print('debug/end shape') return IS