Spaces:
Runtime error
Runtime error
bug fix + cleaned losses
Browse files- code/losses.py +0 -11
code/losses.py
CHANGED
|
@@ -9,9 +9,6 @@ from shapely.geometry import Point
|
|
| 9 |
from shapely.geometry.polygon import Polygon
|
| 10 |
from torchvision import transforms
|
| 11 |
from PIL import Image
|
| 12 |
-
from transformers import CLIPProcessor, CLIPModel
|
| 13 |
-
|
| 14 |
-
from diffusers import StableDiffusionPipeline
|
| 15 |
|
| 16 |
class SDSLoss(nn.Module):
|
| 17 |
def __init__(self, cfg, device, model):
|
|
@@ -20,11 +17,6 @@ class SDSLoss(nn.Module):
|
|
| 20 |
self.device = device
|
| 21 |
self.pipe = model
|
| 22 |
|
| 23 |
-
# self.clip_model = CLIPModel.from_pretrained("openai/clip-vit-large-patch14").to(self.device)
|
| 24 |
-
# self.clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-large-patch14")
|
| 25 |
-
|
| 26 |
-
# default scheduler: PNDMScheduler(beta_start=0.00085, beta_end=0.012,
|
| 27 |
-
# beta_schedule="scaled_linear", num_train_timesteps=1000)
|
| 28 |
self.alphas = self.pipe.scheduler.alphas_cumprod.to(self.device)
|
| 29 |
self.sigmas = (1 - self.pipe.scheduler.alphas_cumprod).to(self.device)
|
| 30 |
|
|
@@ -46,9 +38,6 @@ class SDSLoss(nn.Module):
|
|
| 46 |
|
| 47 |
self.text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
|
| 48 |
self.text_embeddings = self.text_embeddings.repeat_interleave(self.cfg.batch_size, 0)
|
| 49 |
-
del self.pipe.tokenizer
|
| 50 |
-
del self.pipe.text_encoder
|
| 51 |
-
|
| 52 |
|
| 53 |
def forward(self, x_aug):
|
| 54 |
sds_loss = 0
|
|
|
|
| 9 |
from shapely.geometry.polygon import Polygon
|
| 10 |
from torchvision import transforms
|
| 11 |
from PIL import Image
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
class SDSLoss(nn.Module):
|
| 14 |
def __init__(self, cfg, device, model):
|
|
|
|
| 17 |
self.device = device
|
| 18 |
self.pipe = model
|
| 19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
self.alphas = self.pipe.scheduler.alphas_cumprod.to(self.device)
|
| 21 |
self.sigmas = (1 - self.pipe.scheduler.alphas_cumprod).to(self.device)
|
| 22 |
|
|
|
|
| 38 |
|
| 39 |
self.text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
|
| 40 |
self.text_embeddings = self.text_embeddings.repeat_interleave(self.cfg.batch_size, 0)
|
|
|
|
|
|
|
|
|
|
| 41 |
|
| 42 |
def forward(self, x_aug):
|
| 43 |
sds_loss = 0
|