Spaces:
Running
on
Zero
Running
on
Zero
Update inference_coz_single.py
Browse files- inference_coz_single.py +57 -25
inference_coz_single.py
CHANGED
@@ -71,7 +71,7 @@ def _generate_vlm_prompt(
|
|
71 |
return_tensors="pt",
|
72 |
).to(device)
|
73 |
|
74 |
-
# (4) Generate tokens→decode
|
75 |
generated = vlm_model.generate(**inputs, max_new_tokens=128)
|
76 |
# strip off the prompt tokens from each generated sequence:
|
77 |
trimmed = [
|
@@ -86,28 +86,46 @@ def _generate_vlm_prompt(
|
|
86 |
|
87 |
|
88 |
# -------------------------------------------------------------------
|
89 |
-
# Main Function: recursive_multiscale_sr
|
90 |
# -------------------------------------------------------------------
|
91 |
def recursive_multiscale_sr(
|
92 |
input_png_path: str,
|
93 |
upscale: int,
|
94 |
-
|
|
|
|
|
95 |
"""
|
96 |
-
Perform
|
97 |
- input_png_path: path to a single .png file on disk.
|
98 |
- upscale: integer up-scale factor per recursion (e.g. 4).
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
|
|
|
|
|
|
|
|
104 |
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
105 |
###############################
|
106 |
# 1. Fixed hyper-parameters
|
107 |
###############################
|
108 |
device = "cuda"
|
109 |
process_size = 512 # same as args.process_size
|
110 |
-
|
111 |
# model checkpoint paths (hard-coded to your example)
|
112 |
LORA_PATH = "ckpt/SR_LoRA/model_20001.pkl"
|
113 |
VAE_PATH = "ckpt/SR_VAE/vae_encoder_20001.pt"
|
@@ -142,7 +160,7 @@ def recursive_multiscale_sr(
|
|
142 |
###############################
|
143 |
# 3.1 Instantiate the underlying SD3-Euler UNet/VAE/text encoders
|
144 |
sd3 = SD3Euler()
|
145 |
-
# move all text encoders+transformer+VAE to CUDA:
|
146 |
sd3.text_enc_1.to(device)
|
147 |
sd3.text_enc_2.to(device)
|
148 |
sd3.text_enc_3.to(device)
|
@@ -163,7 +181,7 @@ def recursive_multiscale_sr(
|
|
163 |
# (by default, “model_test(...)” takes (lq_tensor, prompt=str) and returns a list[tensor])
|
164 |
|
165 |
###############################
|
166 |
-
# 4. Load the VLM (Qwen2.5-VL)
|
167 |
###############################
|
168 |
vlm_model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
169 |
VLM_NAME,
|
@@ -173,7 +191,7 @@ def recursive_multiscale_sr(
|
|
173 |
vlm_processor = AutoProcessor.from_pretrained(VLM_NAME)
|
174 |
|
175 |
###############################
|
176 |
-
# 5. Pre-allocate a Temporary Directory
|
177 |
# to hold intermediate JPEG/PNG files
|
178 |
###############################
|
179 |
unique_id = uuid.uuid4().hex
|
@@ -193,23 +211,37 @@ def recursive_multiscale_sr(
|
|
193 |
prev_path = os.path.join(td, "step0_prev.png")
|
194 |
img0.save(prev_path)
|
195 |
|
196 |
-
# We will maintain
|
197 |
sr_pil_list: list[Image.Image] = []
|
198 |
-
prompt_list = []
|
199 |
|
200 |
###############################
|
201 |
-
# 7. Recursion loop (
|
202 |
###############################
|
203 |
for rec in range(rec_num):
|
204 |
-
# (A)
|
205 |
prev_pil = Image.open(prev_path).convert("RGB")
|
206 |
-
w, h = prev_pil.size
|
|
|
|
|
207 |
new_w, new_h = w // upscale, h // upscale # e.g. 128×128 for upscale=4
|
208 |
-
|
209 |
-
|
210 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
211 |
right = left + new_w
|
212 |
bottom = top + new_h
|
|
|
213 |
cropped = prev_pil.crop((left, top, right, bottom))
|
214 |
|
215 |
# (B) Resize that crop back up to (512×512) via BICUBIC → zoomed
|
@@ -228,7 +260,7 @@ def recursive_multiscale_sr(
|
|
228 |
)
|
229 |
# (By default, no extra user prompt is appended.)
|
230 |
|
231 |
-
# (D) Prepare the low-res tensor for SR: convert zoomed→Tensor→[0,1]→[−1,1]
|
232 |
to_tensor = transforms.ToTensor()
|
233 |
lq = to_tensor(zoomed).unsqueeze(0).to(device) # shape (1,3,512,512)
|
234 |
lq = (lq * 2.0) - 1.0
|
@@ -252,7 +284,7 @@ def recursive_multiscale_sr(
|
|
252 |
# end for(rec)
|
253 |
|
254 |
###############################
|
255 |
-
# 8. Return the
|
256 |
###############################
|
257 |
-
# The list sr_pil_list = [ SR1, SR2,
|
258 |
-
return sr_pil_list, prompt_list
|
|
|
71 |
return_tensors="pt",
|
72 |
).to(device)
|
73 |
|
74 |
+
# (4) Generate tokens → decode
|
75 |
generated = vlm_model.generate(**inputs, max_new_tokens=128)
|
76 |
# strip off the prompt tokens from each generated sequence:
|
77 |
trimmed = [
|
|
|
86 |
|
87 |
|
88 |
# -------------------------------------------------------------------
|
89 |
+
# Main Function: recursive_multiscale_sr (with multiple centers)
|
90 |
# -------------------------------------------------------------------
|
91 |
def recursive_multiscale_sr(
|
92 |
input_png_path: str,
|
93 |
upscale: int,
|
94 |
+
rec_num: int = 4,
|
95 |
+
centers: list[tuple[float, float]] = None,
|
96 |
+
) -> tuple[list[Image.Image], list[str]]:
|
97 |
"""
|
98 |
+
Perform `rec_num` recursive_multiscale super-resolution steps on a single PNG.
|
99 |
- input_png_path: path to a single .png file on disk.
|
100 |
- upscale: integer up-scale factor per recursion (e.g. 4).
|
101 |
+
- rec_num: how many recursion steps to perform.
|
102 |
+
- centers: a list of normalized (x, y) tuples in [0, 1], one per recursion step,
|
103 |
+
indicating where to center the low-res crop for each step. The list
|
104 |
+
length must equal rec_num. If centers is None, defaults to center=(0.5, 0.5)
|
105 |
+
for all steps.
|
106 |
+
|
107 |
+
Returns a tuple (sr_pil_list, prompt_list), where:
|
108 |
+
- sr_pil_list: list of PIL.Image outputs [SR1, SR2, …, SR_rec_num] in order.
|
109 |
+
- prompt_list: list of the VLM prompts generated at each recursion.
|
110 |
"""
|
111 |
+
###############################
|
112 |
+
# 0. Validate / fill default centers
|
113 |
+
###############################
|
114 |
+
if centers is None:
|
115 |
+
# Default: use center (0.5, 0.5) for every recursion
|
116 |
+
centers = [(0.5, 0.5) for _ in range(rec_num)]
|
117 |
+
else:
|
118 |
+
if not isinstance(centers, (list, tuple)) or len(centers) != rec_num:
|
119 |
+
raise ValueError(
|
120 |
+
f"`centers` must be a list of {rec_num} (x,y) tuples, but got length {len(centers)}."
|
121 |
+
)
|
122 |
+
|
123 |
###############################
|
124 |
# 1. Fixed hyper-parameters
|
125 |
###############################
|
126 |
device = "cuda"
|
127 |
process_size = 512 # same as args.process_size
|
128 |
+
|
129 |
# model checkpoint paths (hard-coded to your example)
|
130 |
LORA_PATH = "ckpt/SR_LoRA/model_20001.pkl"
|
131 |
VAE_PATH = "ckpt/SR_VAE/vae_encoder_20001.pt"
|
|
|
160 |
###############################
|
161 |
# 3.1 Instantiate the underlying SD3-Euler UNet/VAE/text encoders
|
162 |
sd3 = SD3Euler()
|
163 |
+
# move all text encoders + transformer + VAE to CUDA:
|
164 |
sd3.text_enc_1.to(device)
|
165 |
sd3.text_enc_2.to(device)
|
166 |
sd3.text_enc_3.to(device)
|
|
|
181 |
# (by default, “model_test(...)” takes (lq_tensor, prompt=str) and returns a list[tensor])
|
182 |
|
183 |
###############################
|
184 |
+
# 4. Load the VLM (Qwen2.5-VL)
|
185 |
###############################
|
186 |
vlm_model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
187 |
VLM_NAME,
|
|
|
191 |
vlm_processor = AutoProcessor.from_pretrained(VLM_NAME)
|
192 |
|
193 |
###############################
|
194 |
+
# 5. Pre-allocate a Temporary Directory
|
195 |
# to hold intermediate JPEG/PNG files
|
196 |
###############################
|
197 |
unique_id = uuid.uuid4().hex
|
|
|
211 |
prev_path = os.path.join(td, "step0_prev.png")
|
212 |
img0.save(prev_path)
|
213 |
|
214 |
+
# We will maintain lists of PIL outputs and prompts:
|
215 |
sr_pil_list: list[Image.Image] = []
|
216 |
+
prompt_list: list[str] = []
|
217 |
|
218 |
###############################
|
219 |
+
# 7. Recursion loop (now up to rec_num times)
|
220 |
###############################
|
221 |
for rec in range(rec_num):
|
222 |
+
# (A) Load the previous SR output (or original) and compute crop window
|
223 |
prev_pil = Image.open(prev_path).convert("RGB")
|
224 |
+
w, h = prev_pil.size # should be (512×512) each time
|
225 |
+
|
226 |
+
# (1) Compute the “low-res” window size:
|
227 |
new_w, new_h = w // upscale, h // upscale # e.g. 128×128 for upscale=4
|
228 |
+
|
229 |
+
# (2) Map normalized center → pixel center, then clamp so crop stays in bounds:
|
230 |
+
cx_norm, cy_norm = centers[rec]
|
231 |
+
cx = int(cx_norm * w)
|
232 |
+
cy = int(cy_norm * h)
|
233 |
+
half_w = new_w // 2
|
234 |
+
half_h = new_h // 2
|
235 |
+
|
236 |
+
# If center in pixels is too close to left/top, clamp so left=0 or top=0; same on right/bottom
|
237 |
+
left = cx - half_w
|
238 |
+
top = cy - half_h
|
239 |
+
# clamp left ∈ [0, w - new_w], top ∈ [0, h - new_h]
|
240 |
+
left = max(0, min(left, w - new_w))
|
241 |
+
top = max(0, min(top, h - new_h))
|
242 |
right = left + new_w
|
243 |
bottom = top + new_h
|
244 |
+
|
245 |
cropped = prev_pil.crop((left, top, right, bottom))
|
246 |
|
247 |
# (B) Resize that crop back up to (512×512) via BICUBIC → zoomed
|
|
|
260 |
)
|
261 |
# (By default, no extra user prompt is appended.)
|
262 |
|
263 |
+
# (D) Prepare the low-res tensor for SR: convert zoomed → Tensor → [0,1] → [−1,1]
|
264 |
to_tensor = transforms.ToTensor()
|
265 |
lq = to_tensor(zoomed).unsqueeze(0).to(device) # shape (1,3,512,512)
|
266 |
lq = (lq * 2.0) - 1.0
|
|
|
284 |
# end for(rec)
|
285 |
|
286 |
###############################
|
287 |
+
# 8. Return the SR outputs & prompts
|
288 |
###############################
|
289 |
+
# The list sr_pil_list = [ SR1, SR2, …, SR_rec_num ] in order.
|
290 |
+
return sr_pil_list, prompt_list
|