Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 17:24:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

jumbo merge

Browse Source
This commit is contained in:
Vladimir Mandic
2023-06-13 11:58:14 -04:00
parent 0d101b9def
commit cb307399dd
72 changed files with 535 additions and 408 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/models/diffusion/ddpm_edit.py
+18 -25
View File
@@ -52,7 +52,7 @@ class DDPM(pl.LightningModule):
beta_schedule="linear",
loss_type="l2",
ckpt_path=None,
ignore_keys=[],
ignore_keys=None,
load_only_unet=False,
monitor="val/loss",
use_ema=True,
@@ -107,7 +107,7 @@ class DDPM(pl.LightningModule):
print(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.")
if ckpt_path is not None:
self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys, only_model=load_only_unet)
self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys or [], only_model=load_only_unet)
# If initialing from EMA-only checkpoint, create EMA model after loading.
if self.use_ema and not load_ema:
@@ -194,7 +194,8 @@ class DDPM(pl.LightningModule):
if context is not None:
print(f"{context}: Restored training weights")
def init_from_ckpt(self, path, ignore_keys=list(), only_model=False):
def init_from_ckpt(self, path, ignore_keys=None, only_model=False):
ignore_keys = ignore_keys or []
sd = torch.load(path, map_location="cpu")
if "state_dict" in list(sd.keys()):
sd = sd["state_dict"]
@@ -403,7 +404,7 @@ class DDPM(pl.LightningModule):
@torch.no_grad()
def log_images(self, batch, N=8, n_row=2, sample=True, return_keys=None, **kwargs):
log = dict()
log = {}
x = self.get_input(batch, self.first_stage_key)
N = min(x.shape[0], N)
n_row = min(x.shape[0], n_row)
@@ -411,7 +412,7 @@ class DDPM(pl.LightningModule):
log["inputs"] = x
# get diffusion row
diffusion_row = list()
diffusion_row = []
x_start = x[:n_row]
for t in range(self.num_timesteps):
@@ -473,13 +474,13 @@ class LatentDiffusion(DDPM):
conditioning_key = None
ckpt_path = kwargs.pop("ckpt_path", None)
ignore_keys = kwargs.pop("ignore_keys", [])
super().__init__(conditioning_key=conditioning_key, *args, load_ema=load_ema, **kwargs)
super().__init__(*args, conditioning_key=conditioning_key, load_ema=load_ema, **kwargs)
self.concat_mode = concat_mode
self.cond_stage_trainable = cond_stage_trainable
self.cond_stage_key = cond_stage_key
try:
self.num_downs = len(first_stage_config.params.ddconfig.ch_mult) - 1
except:
except Exception:
self.num_downs = 0
if not scale_by_std:
self.scale_factor = scale_factor
@@ -891,16 +892,6 @@ class LatentDiffusion(DDPM):
c = self.q_sample(x_start=c, t=tc, noise=torch.randn_like(c.float()))
return self.p_losses(x, c, t, *args, **kwargs)
def _rescale_annotations(self, bboxes, crop_coordinates): # TODO: move to dataset
def rescale_bbox(bbox):
x0 = clamp((bbox[0] - crop_coordinates[0]) / crop_coordinates[2])
y0 = clamp((bbox[1] - crop_coordinates[1]) / crop_coordinates[3])
w = min(bbox[2] / crop_coordinates[2], 1 - x0)
h = min(bbox[3] / crop_coordinates[3], 1 - y0)
return x0, y0, w, h
return [rescale_bbox(b) for b in bboxes]
def apply_model(self, x_noisy, t, cond, return_ids=False):
if isinstance(cond, dict):
@@ -1140,7 +1131,7 @@ class LatentDiffusion(DDPM):
if cond is not None:
if isinstance(cond, dict):
cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else
list(map(lambda x: x[:batch_size], cond[key])) for key in cond}
[x[:batch_size] for x in cond[key]] for key in cond}
else:
cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size]
@@ -1171,8 +1162,10 @@ class LatentDiffusion(DDPM):
if i % log_every_t == 0 or i == timesteps - 1:
intermediates.append(x0_partial)
if callback: callback(i)
if img_callback: img_callback(img, i)
if callback:
callback(i)
if img_callback:
img_callback(img, i)
return img, intermediates
@torch.no_grad()
@@ -1235,7 +1228,7 @@ class LatentDiffusion(DDPM):
if cond is not None:
if isinstance(cond, dict):
cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else
list(map(lambda x: x[:batch_size], cond[key])) for key in cond}
[x[:batch_size] for x in cond[key]] for key in cond}
else:
cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size]
return self.p_sample_loop(cond,
@@ -1267,7 +1260,7 @@ class LatentDiffusion(DDPM):
use_ddim = False
log = dict()
log = {}
z, c, x, xrec, xc = self.get_input(batch, self.first_stage_key,
return_first_stage_outputs=True,
force_c_encode=True,
@@ -1337,7 +1330,7 @@ class LatentDiffusion(DDPM):
if inpaint:
# make a simple center square
b, h, w = z.shape[0], z.shape[2], z.shape[3]
h, w = z.shape[2], z.shape[3]
mask = torch.ones(N, h, w).to(self.device)
# zeros will be filled in
mask[:, h // 4:3 * h // 4, w // 4:3 * w // 4] = 0.
@@ -1439,10 +1432,10 @@ class Layout2ImgDiffusion(LatentDiffusion):
# TODO: move all layout-specific hacks to this class
def __init__(self, cond_stage_key, *args, **kwargs):
assert cond_stage_key == 'coordinates_bbox', 'Layout2ImgDiffusion only for cond_stage_key="coordinates_bbox"'
super().__init__(cond_stage_key=cond_stage_key, *args, **kwargs)
super().__init__(*args, cond_stage_key=cond_stage_key, **kwargs)
def log_images(self, batch, N=8, *args, **kwargs):
logs = super().log_images(batch=batch, N=N, *args, **kwargs)
logs = super().log_images(*args, batch=batch, N=N, **kwargs)
key = 'train' if self.training else 'validation'
dset = self.trainer.datamodule.datasets[key]