fix no eval bug (#790)

ppfleetx/data/__init__.py

@@ -58,6 +58,8 @@ def build_dataset(config, mode):
 
 def build_dataloader(config, mode):
     dataset = build_dataset(config, mode)
+    if dataset is None:
+        return None
 
     batch_sampler = None
     # build sampler

ppfleetx/data/dataset/multimodal_dataset.py

@@ -78,11 +78,11 @@ def data_augmentation_for_imagen(img, resolution):
     arr = deepcopy(img)
     while min(*arr.size) >= 2 * resolution:
         arr = arr.resize(
-            tuple(x // 2 for x in arr.size), resample=Image.Resampling.BOX)
+            tuple(x // 2 for x in arr.size), resample=Image.BOX)
     scale = resolution / min(*arr.size)
     arr = arr.resize(
         tuple(round(x * scale) for x in arr.size),
-        resample=Image.Resampling.BICUBIC)
+        resample=Image.BICUBIC)
 
     arr = np.array(arr.convert("RGB"))
     crop_y = (arr.shape[0] - resolution) // 2

ppfleetx/models/multimodal_model/imagen/modeling.py

@@ -21,8 +21,8 @@
 import paddle.vision.transforms as T
 
 from .unet import Unet
-from .utils import (GaussianDiffusionContinuousTimes, default, exists,
-                    cast_tuple, first, maybe, eval_decorator, identity,
+from .utils import (GaussianDiffusionContinuousTimes, default, cast_tuple,
+                    first, maybe, eval_decorator, identity,
                     pad_tuple_to_length, right_pad_dims_to, resize_image_to,
                     normalize_neg_one_to_one, rearrange, repeat, reduce,
                     unnormalize_zero_to_one, cast_uint8_images_to_float)
@@ -195,9 +195,8 @@ def __init__(self,
         # randomly cropping for upsampler training
 
         self.random_crop_sizes = cast_tuple(random_crop_sizes, num_unets)
-        assert not exists(
-            first(self.random_crop_sizes)
-        ), 'you should not need to randomly crop image during training for base unet, only for upsamplers - so pass in `random_crop_sizes = (None, 128, 256)` as example'
+        assert first(
+            self.random_crop_sizes) is None, 'you should not need to randomly crop image during training for base unet, only for upsamplers - so pass in `random_crop_sizes = (None, 128, 256)` as example'
         # lowres augmentation noise schedule
 
         self.lowres_noise_schedule = GaussianDiffusionContinuousTimes(
@@ -284,22 +283,17 @@ def get_unet(self, unet_number):
         assert 0 < unet_number <= len(self.unets)
         index = unet_number - 1
 
-        if isinstance(self.unets, nn.LayerList):
-            unets_list = [unet for unet in self.unets]
-            delattr(self, 'unets')
-            self.unets = unets_list
         self.unet_being_trained_index = index
         return self.unets[index]
 
     def reset_unets(self, ):
-        self.unets = nn.LayerList([*self.unets])
         self.unet_being_trained_index = -1
 
     @contextmanager
     def one_unet_in_gpu(self, unet_number=None, unet=None):
-        assert exists(unet_number) ^ exists(unet)
+        assert (unet_number is not None) ^ (unet is not None)
 
-        if exists(unet_number):
+        if unet_number is not None:
             unet = self.unets[unet_number - 1]
 
         yield
@@ -320,7 +314,6 @@ def p_mean_variance(self,
                         unet,
                         x,
                         t,
-                        *,
                         noise_scheduler,
                         text_embeds=None,
                         text_mask=None,
@@ -370,7 +363,6 @@ def p_sample(self,
                  unet,
                  x,
                  t,
-                 *,
                  noise_scheduler,
                  t_next=None,
                  text_embeds=None,
@@ -412,7 +404,6 @@ def p_sample(self,
     def p_sample_loop(self,
                       unet,
                       shape,
-                      *,
                       noise_scheduler,
                       lowres_cond_img=None,
                       lowres_noise_times=None,
@@ -433,7 +424,7 @@ def p_sample_loop(self,
 
         # prepare inpainting
 
-        has_inpainting = exists(inpaint_images) and exists(inpaint_masks)
+        has_inpainting = inpaint_images is not None and inpaint_masks is not None
         resample_times = inpaint_resample_times if has_inpainting else 1
 
         if has_inpainting:
@@ -532,18 +523,18 @@ def sample(
             batch_size = text_embeds.shape[0]
 
         assert not (
-            self.condition_on_text and not exists(text_embeds)
+            self.condition_on_text and text_embeds is None
         ), 'text or text encodings must be passed into imagen if specified'
         assert not (
-            not self.condition_on_text and exists(text_embeds)
+            not self.condition_on_text and text_embeds is not None
         ), 'imagen specified not to be conditioned on text, yet it is presented'
         assert not (
-            exists(text_embeds) and
+            text_embeds is not None and
             text_embeds.shape[-1] != self.text_embed_dim
         ), f'invalid text embedding dimension being passed in (should be {self.text_embed_dim})'
 
         assert not (
-            exists(inpaint_images) ^ exists(inpaint_masks)
+            (inpaint_images is not None) ^ (inpaint_masks is not None)
         ), 'inpaint images and masks must be both passed in to do inpainting'
 
         outputs = []
@@ -609,8 +600,7 @@ def sample(
 
                 outputs.append(img)
 
-            if exists(stop_at_unet_number
-                      ) and stop_at_unet_number == unet_number:
+            if stop_at_unet_number is not None and stop_at_unet_number == unet_number:
                 break
 
         output_index = -1 if not return_all_unet_outputs else slice(
@@ -633,7 +623,6 @@ def p_losses(self,
                  unet,
                  x_start,
                  times,
-                 *,
                  noise_scheduler,
                  lowres_cond_img=None,
                  lowres_aug_times=None,
@@ -655,7 +644,7 @@ def p_losses(self,
         # random cropping during training
         # for upsamplers
 
-        if exists(random_crop_size):
+        if random_crop_size is not None:
             aug = K.RandomCrop((random_crop_size, random_crop_size), p=1.)
             # make sure low res conditioner and image both get augmented the same way
             # detailed https://kornia.readthedocs.io/en/latest/augmentation.module.html?highlight=randomcrop#kornia.augmentation.RandomCrop
@@ -672,7 +661,7 @@ def p_losses(self,
         # at sample time, they then fix the noise level of 0.1 - 0.3
 
         lowres_cond_img_noisy = None
-        if exists(lowres_cond_img):
+        if lowres_cond_img is not None:
             lowres_aug_times = default(lowres_aug_times, times)
             lowres_cond_img_noisy, _ = self.lowres_noise_schedule.q_sample(
                 x_start=lowres_cond_img,
@@ -715,11 +704,10 @@ def forward(self,
         assert images.shape[-1] == images.shape[
             -2], f'the images you pass in must be a square, but received dimensions of {images.shape[2]}, {images.shape[-1]}'
         assert not (
-            len(self.unets) > 1 and not exists(unet_number)
+            len(self.unets) > 1 and unet_number is None
         ), f'you must specify which unet you want trained, from a range of 1 to {len(self.unets)}, if you are training cascading DDPM (multiple unets)'
         unet_number = default(unet_number, 1)
-        assert not exists(
-            self.only_train_unet_number
+        assert (self.only_train_unet_number is None
         ) or self.only_train_unet_number == unet_number, 'you can only train on unet #{self.only_train_unet_number}'
 
         images = cast_uint8_images_to_float(images)
@@ -748,19 +736,19 @@ def forward(self,
                 text_masks, lambda: paddle.any(text_embeds != 0., axis=-1))
 
         assert not (
-            self.condition_on_text and not exists(text_embeds)
+            self.condition_on_text and text_embeds is None
         ), 'text or text encodings must be passed into decoder if specified'
         assert not (
-            not self.condition_on_text and exists(text_embeds)
+            not self.condition_on_text and text_embeds is not None
         ), 'decoder specified not to be conditioned on text, yet it is presented'
 
         assert not (
-            exists(text_embeds) and
+            (text_embeds is not None) and
             text_embeds.shape[-1] != self.text_embed_dim
         ), f'invalid text embedding dimension being passed in (should be {self.text_embed_dim})'
 
         lowres_cond_img = lowres_aug_times = None
-        if exists(prev_image_size):
+        if prev_image_size is not None:
             lowres_cond_img = resize_image_to(
                 images, prev_image_size, clamp_range=self.input_image_range)
             lowres_cond_img = resize_image_to(