Merge remote-tracking branch 'origin/main'

scripts/img2img.py

@@ -1,6 +1,7 @@
 """make variations of input image"""
 
 import argparse, os, sys, glob
+import PIL
 import torch
 import numpy as np
 from omegaconf import OmegaConf
@@ -9,6 +10,8 @@ from tqdm import tqdm, trange
 from itertools import islice
 from einops import rearrange, repeat
 from torchvision.utils import make_grid
+from torch import autocast
+from contextlib import nullcontext
 import time
 from pytorch_lightning import seed_everything
 
@@ -43,8 +46,12 @@ def load_model_from_config(config, ckpt, verbose=False):
 
 
 def load_img(path):
-    image = np.array(Image.open(path).convert("RGB"))
+    image = Image.open(path).convert("RGB")
-    image = image.astype(np.float32) / 255.0
+    w, h = image.size
+    print(f"loaded input image of size ({w}, {h}) from {path}")
+    w, h = map(lambda x: x - x % 32, (w, h))  # resize to integer multiple of 32
+    image = image.resize((w, h), resample=PIL.Image.LANCZOS)
+    image = np.array(image).astype(np.float32) / 255.0
     image = image[None].transpose(0, 3, 1, 2)
     image = torch.from_numpy(image)
     return 2.*image - 1.
@@ -119,20 +126,6 @@ def main():
         help="sample this often",
     )
 
-    parser.add_argument(
-        "--H",
-        type=int,
-        default=256,
-        help="image height, in pixel space",
-    )
-
-    parser.add_argument(
-        "--W",
-        type=int,
-        default=256,
-        help="image width, in pixel space",
-    )
-
     parser.add_argument(
         "--C",
         type=int,
@@ -149,7 +142,7 @@ def main():
     parser.add_argument(
         "--n_samples",
         type=int,
-        default=8,
+        default=2,
         help="how many samples to produce for each given prompt. A.k.a batch size",
     )
 
@@ -170,7 +163,7 @@ def main():
     parser.add_argument(
         "--strength",
         type=float,
-        default=0.3,
+        default=0.75,
         help="strength for noising/unnoising. 1.0 corresponds to full destruction of information in init image",
     )
 
@@ -197,6 +190,14 @@ def main():
         default=42,
         help="the seed (for reproducible sampling)",
     )
+    parser.add_argument(
+        "--precision",
+        type=str,
+        help="evaluate at this precision",
+        choices=["full", "autocast"],
+        default="autocast"
+    )
+
     opt = parser.parse_args()
     seed_everything(opt.seed)
 
@@ -244,7 +245,9 @@ def main():
     t_enc = int(opt.strength * opt.ddim_steps)
     print(f"target t_enc is {t_enc} steps")
 
+    precision_scope = autocast if opt.precision == "autocast" else nullcontext
     with torch.no_grad():
+        with precision_scope("cuda"):
             with model.ema_scope():
                 tic = time.time()
                 all_samples = list()
@@ -288,7 +291,7 @@ def main():
                 toc = time.time()
 
     print(f"Your samples are ready and waiting for you here: \n{outpath} \n"
-          f"Sampling took {toc - tic}s, i.e. produced {opt.n_iter * opt.n_samples / (toc - tic):.2f} samples/sec."
+          f"Sampling took {toc - tic}s, i.e., produced {opt.n_iter * opt.n_samples / (toc - tic):.2f} samples/sec."
           f" \nEnjoy.")
 
 

scripts/prune-ckpt.py

@@ -3,7 +3,7 @@ import torch
 import fire
 
 
-def prune_it(p):
+def prune_it(p, keep_only_ema=False):
     print(f"prunin' in path: {p}")
     size_initial = os.path.getsize(p)
     nsd = dict()
@@ -16,12 +16,30 @@ def prune_it(p):
         print(f"removing optimizer states for path {p}")
     if "global_step" in sd:
         print(f"This is global step {sd['global_step']}.")
-    fn = f"{os.path.splitext(p)[0]}-pruned.ckpt"
+    if keep_only_ema:
+        sd = nsd["state_dict"].copy()
+        # infer ema keys
+        ema_keys = {k: "model_ema." + k[6:].replace(".", "") for k in sd.keys() if k.startswith("model.")}
+        new_sd = dict()
+
+        for k in sd:
+            if k in ema_keys:
+                new_sd[k] = sd[ema_keys[k]]
+            elif not k.startswith("model_ema.") or k in ["model_ema.num_updates", "model_ema.decay"]:
+                new_sd[k] = sd[k]
+
+        assert len(new_sd) == len(sd) - len(ema_keys)
+        nsd["state_dict"] = new_sd
+
+    fn = f"{os.path.splitext(p)[0]}-pruned.ckpt" if not keep_only_ema else f"{os.path.splitext(p)[0]}-ema-pruned.ckpt"
     print(f"saving pruned checkpoint at: {fn}")
     torch.save(nsd, fn)
     newsize = os.path.getsize(fn)
-    print(f"New ckpt size: {newsize*1e-9:.2f} GB. "
+    MSG = f"New ckpt size: {newsize*1e-9:.2f} GB. " + \
-          f"Saved {(size_initial - newsize)*1e-9:.2f} GB by removing optimizer states")
+          f"Saved {(size_initial - newsize)*1e-9:.2f} GB by removing optimizer states"
+    if keep_only_ema:
+        MSG += " and non-EMA weights"
+    print(MSG)
 
 
 if __name__ == "__main__":

scripts/txt2img.py

@@ -9,6 +9,8 @@ from einops import rearrange
 from torchvision.utils import make_grid
 import time
 from pytorch_lightning import seed_everything
+from torch import autocast
+from contextlib import contextmanager, nullcontext
 
 from ldm.util import instantiate_from_config
 from ldm.models.diffusion.ddim import DDIMSampler
@@ -178,6 +180,13 @@ def main():
         default=42,
         help="the seed (for reproducible sampling)",
     )
+    parser.add_argument(
+        "--precision",
+        type=str,
+        help="evaluate at this precision",
+        choices=["full", "autocast"],
+        default="autocast"
+    )
     opt = parser.parse_args()
     seed_everything(opt.seed)
 
@@ -217,7 +226,9 @@ def main():
     if opt.fixed_code:
         start_code = torch.randn([opt.n_samples, opt.C, opt.H // opt.f, opt.W // opt.f], device=device)
 
+    precision_scope = autocast if opt.precision=="autocast" else nullcontext
     with torch.no_grad():
+        with precision_scope("cuda"):
             with model.ema_scope():
                 tic = time.time()
                 all_samples = list()