Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
3838dc1
1
Parent(s):
7c2f6e2
[Release] v1.0.1
Browse files- improve the performance
- improve efficiency
- app.py +23 -15
- depthcrafter/utils.py +44 -0
- run.py +2 -50
app.py
CHANGED
|
@@ -17,11 +17,11 @@ from huggingface_hub import hf_hub_download
|
|
| 17 |
from depthcrafter.utils import read_video_frames, vis_sequence_depth, save_video
|
| 18 |
|
| 19 |
examples = [
|
| 20 |
-
["examples/example_01.mp4",
|
| 21 |
-
["examples/example_02.mp4",
|
| 22 |
-
["examples/example_03.mp4",
|
| 23 |
-
["examples/example_04.mp4",
|
| 24 |
-
["examples/example_05.mp4",
|
| 25 |
]
|
| 26 |
|
| 27 |
|
|
@@ -39,18 +39,18 @@ pipe = DepthCrafterPipeline.from_pretrained(
|
|
| 39 |
pipe.to("cuda")
|
| 40 |
|
| 41 |
|
| 42 |
-
@spaces.GPU(duration=
|
| 43 |
def infer_depth(
|
| 44 |
video: str,
|
| 45 |
num_denoising_steps: int,
|
| 46 |
guidance_scale: float,
|
| 47 |
max_res: int = 1024,
|
| 48 |
-
process_length: int =
|
|
|
|
| 49 |
#
|
| 50 |
save_folder: str = "./demo_output",
|
| 51 |
window_size: int = 110,
|
| 52 |
overlap: int = 25,
|
| 53 |
-
target_fps: int = 15,
|
| 54 |
seed: int = 42,
|
| 55 |
track_time: bool = True,
|
| 56 |
save_npz: bool = False,
|
|
@@ -59,7 +59,6 @@ def infer_depth(
|
|
| 59 |
pipe.enable_xformers_memory_efficient_attention()
|
| 60 |
|
| 61 |
frames, target_fps = read_video_frames(video, process_length, target_fps, max_res)
|
| 62 |
-
print(f"==> video name: {video}, frames shape: {frames.shape}")
|
| 63 |
|
| 64 |
# inference the depth map using the DepthCrafter pipeline
|
| 65 |
with torch.inference_mode():
|
|
@@ -82,6 +81,7 @@ def infer_depth(
|
|
| 82 |
vis = vis_sequence_depth(res)
|
| 83 |
# save the depth map and visualization with the target FPS
|
| 84 |
save_path = os.path.join(save_folder, os.path.splitext(os.path.basename(video))[0])
|
|
|
|
| 85 |
os.makedirs(os.path.dirname(save_path), exist_ok=True)
|
| 86 |
if save_npz:
|
| 87 |
np.savez_compressed(save_path + ".npz", depth=res)
|
|
@@ -155,14 +155,14 @@ def construct_demo():
|
|
| 155 |
label="num denoising steps",
|
| 156 |
minimum=1,
|
| 157 |
maximum=25,
|
| 158 |
-
value=
|
| 159 |
step=1,
|
| 160 |
)
|
| 161 |
guidance_scale = gr.Slider(
|
| 162 |
label="cfg scale",
|
| 163 |
minimum=1.0,
|
| 164 |
maximum=1.2,
|
| 165 |
-
value=1.
|
| 166 |
step=0.1,
|
| 167 |
)
|
| 168 |
max_res = gr.Slider(
|
|
@@ -174,11 +174,18 @@ def construct_demo():
|
|
| 174 |
)
|
| 175 |
process_length = gr.Slider(
|
| 176 |
label="process length",
|
| 177 |
-
minimum
|
| 178 |
maximum=280,
|
| 179 |
value=60,
|
| 180 |
step=1,
|
| 181 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 182 |
generate_btn = gr.Button("Generate")
|
| 183 |
with gr.Column(scale=2):
|
| 184 |
pass
|
|
@@ -191,6 +198,7 @@ def construct_demo():
|
|
| 191 |
guidance_scale,
|
| 192 |
max_res,
|
| 193 |
process_length,
|
|
|
|
| 194 |
],
|
| 195 |
outputs=[output_video_1, output_video_2],
|
| 196 |
fn=infer_depth,
|
|
@@ -216,6 +224,7 @@ def construct_demo():
|
|
| 216 |
guidance_scale,
|
| 217 |
max_res,
|
| 218 |
process_length,
|
|
|
|
| 219 |
],
|
| 220 |
outputs=[output_video_1, output_video_2],
|
| 221 |
)
|
|
@@ -223,9 +232,8 @@ def construct_demo():
|
|
| 223 |
return depthcrafter_iface
|
| 224 |
|
| 225 |
|
| 226 |
-
demo = construct_demo()
|
| 227 |
-
|
| 228 |
if __name__ == "__main__":
|
|
|
|
| 229 |
demo.queue()
|
| 230 |
-
# demo.launch(server_name="0.0.0.0", server_port=
|
| 231 |
demo.launch(share=True)
|
|
|
|
| 17 |
from depthcrafter.utils import read_video_frames, vis_sequence_depth, save_video
|
| 18 |
|
| 19 |
examples = [
|
| 20 |
+
["examples/example_01.mp4", 5, 1.0, 1024, -1, -1],
|
| 21 |
+
["examples/example_02.mp4", 5, 1.0, 1024, -1, -1],
|
| 22 |
+
["examples/example_03.mp4", 5, 1.0, 1024, -1, -1],
|
| 23 |
+
["examples/example_04.mp4", 5, 1.0, 1024, -1, -1],
|
| 24 |
+
["examples/example_05.mp4", 5, 1.0, 1024, -1, -1],
|
| 25 |
]
|
| 26 |
|
| 27 |
|
|
|
|
| 39 |
pipe.to("cuda")
|
| 40 |
|
| 41 |
|
| 42 |
+
@spaces.GPU(duration=120)
|
| 43 |
def infer_depth(
|
| 44 |
video: str,
|
| 45 |
num_denoising_steps: int,
|
| 46 |
guidance_scale: float,
|
| 47 |
max_res: int = 1024,
|
| 48 |
+
process_length: int = -1,
|
| 49 |
+
target_fps: int = -1,
|
| 50 |
#
|
| 51 |
save_folder: str = "./demo_output",
|
| 52 |
window_size: int = 110,
|
| 53 |
overlap: int = 25,
|
|
|
|
| 54 |
seed: int = 42,
|
| 55 |
track_time: bool = True,
|
| 56 |
save_npz: bool = False,
|
|
|
|
| 59 |
pipe.enable_xformers_memory_efficient_attention()
|
| 60 |
|
| 61 |
frames, target_fps = read_video_frames(video, process_length, target_fps, max_res)
|
|
|
|
| 62 |
|
| 63 |
# inference the depth map using the DepthCrafter pipeline
|
| 64 |
with torch.inference_mode():
|
|
|
|
| 81 |
vis = vis_sequence_depth(res)
|
| 82 |
# save the depth map and visualization with the target FPS
|
| 83 |
save_path = os.path.join(save_folder, os.path.splitext(os.path.basename(video))[0])
|
| 84 |
+
print(f"==> saving results to {save_path}")
|
| 85 |
os.makedirs(os.path.dirname(save_path), exist_ok=True)
|
| 86 |
if save_npz:
|
| 87 |
np.savez_compressed(save_path + ".npz", depth=res)
|
|
|
|
| 155 |
label="num denoising steps",
|
| 156 |
minimum=1,
|
| 157 |
maximum=25,
|
| 158 |
+
value=5,
|
| 159 |
step=1,
|
| 160 |
)
|
| 161 |
guidance_scale = gr.Slider(
|
| 162 |
label="cfg scale",
|
| 163 |
minimum=1.0,
|
| 164 |
maximum=1.2,
|
| 165 |
+
value=1.0,
|
| 166 |
step=0.1,
|
| 167 |
)
|
| 168 |
max_res = gr.Slider(
|
|
|
|
| 174 |
)
|
| 175 |
process_length = gr.Slider(
|
| 176 |
label="process length",
|
| 177 |
+
minimum=-1,
|
| 178 |
maximum=280,
|
| 179 |
value=60,
|
| 180 |
step=1,
|
| 181 |
)
|
| 182 |
+
process_target_fps = gr.Slider(
|
| 183 |
+
label="target FPS",
|
| 184 |
+
minimum=-1,
|
| 185 |
+
maximum=30,
|
| 186 |
+
value=15,
|
| 187 |
+
step=1,
|
| 188 |
+
)
|
| 189 |
generate_btn = gr.Button("Generate")
|
| 190 |
with gr.Column(scale=2):
|
| 191 |
pass
|
|
|
|
| 198 |
guidance_scale,
|
| 199 |
max_res,
|
| 200 |
process_length,
|
| 201 |
+
process_target_fps,
|
| 202 |
],
|
| 203 |
outputs=[output_video_1, output_video_2],
|
| 204 |
fn=infer_depth,
|
|
|
|
| 224 |
guidance_scale,
|
| 225 |
max_res,
|
| 226 |
process_length,
|
| 227 |
+
process_target_fps,
|
| 228 |
],
|
| 229 |
outputs=[output_video_1, output_video_2],
|
| 230 |
)
|
|
|
|
| 232 |
return depthcrafter_iface
|
| 233 |
|
| 234 |
|
|
|
|
|
|
|
| 235 |
if __name__ == "__main__":
|
| 236 |
+
demo = construct_demo()
|
| 237 |
demo.queue()
|
| 238 |
+
# demo.launch(server_name="0.0.0.0", server_port=12345, debug=True, share=False)
|
| 239 |
demo.launch(share=True)
|
depthcrafter/utils.py
CHANGED
|
@@ -5,6 +5,50 @@ import PIL.Image
|
|
| 5 |
import matplotlib.cm as cm
|
| 6 |
import mediapy
|
| 7 |
import torch
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
|
| 10 |
def save_video(
|
|
|
|
| 5 |
import matplotlib.cm as cm
|
| 6 |
import mediapy
|
| 7 |
import torch
|
| 8 |
+
from decord import VideoReader, cpu
|
| 9 |
+
|
| 10 |
+
dataset_res_dict = {
|
| 11 |
+
"sintel": [448, 1024],
|
| 12 |
+
"scannet": [640, 832],
|
| 13 |
+
"KITTI": [384, 1280],
|
| 14 |
+
"bonn": [512, 640],
|
| 15 |
+
"NYUv2": [448, 640],
|
| 16 |
+
}
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
def read_video_frames(video_path, process_length, target_fps, max_res, dataset="open"):
|
| 20 |
+
if dataset == "open":
|
| 21 |
+
print("==> processing video: ", video_path)
|
| 22 |
+
vid = VideoReader(video_path, ctx=cpu(0))
|
| 23 |
+
print("==> original video shape: ", (len(vid), *vid.get_batch([0]).shape[1:]))
|
| 24 |
+
original_height, original_width = vid.get_batch([0]).shape[1:3]
|
| 25 |
+
height = round(original_height / 64) * 64
|
| 26 |
+
width = round(original_width / 64) * 64
|
| 27 |
+
if max(height, width) > max_res:
|
| 28 |
+
scale = max_res / max(original_height, original_width)
|
| 29 |
+
height = round(original_height * scale / 64) * 64
|
| 30 |
+
width = round(original_width * scale / 64) * 64
|
| 31 |
+
else:
|
| 32 |
+
height = dataset_res_dict[dataset][0]
|
| 33 |
+
width = dataset_res_dict[dataset][1]
|
| 34 |
+
|
| 35 |
+
vid = VideoReader(video_path, ctx=cpu(0), width=width, height=height)
|
| 36 |
+
|
| 37 |
+
fps = vid.get_avg_fps() if target_fps == -1 else target_fps
|
| 38 |
+
stride = round(vid.get_avg_fps() / fps)
|
| 39 |
+
stride = max(stride, 1)
|
| 40 |
+
frames_idx = list(range(0, len(vid), stride))
|
| 41 |
+
print(
|
| 42 |
+
f"==> downsampled shape: {len(frames_idx), *vid.get_batch([0]).shape[1:]}, with stride: {stride}"
|
| 43 |
+
)
|
| 44 |
+
if process_length != -1 and process_length < len(frames_idx):
|
| 45 |
+
frames_idx = frames_idx[:process_length]
|
| 46 |
+
print(
|
| 47 |
+
f"==> final processing shape: {len(frames_idx), *vid.get_batch([0]).shape[1:]}"
|
| 48 |
+
)
|
| 49 |
+
frames = vid.get_batch(frames_idx).asnumpy().astype("float32") / 255.0
|
| 50 |
+
|
| 51 |
+
return frames, fps
|
| 52 |
|
| 53 |
|
| 54 |
def save_video(
|
run.py
CHANGED
|
@@ -3,21 +3,12 @@ import os
|
|
| 3 |
import numpy as np
|
| 4 |
import torch
|
| 5 |
|
| 6 |
-
from decord import VideoReader, cpu
|
| 7 |
from diffusers.training_utils import set_seed
|
| 8 |
from fire import Fire
|
| 9 |
|
| 10 |
from depthcrafter.depth_crafter_ppl import DepthCrafterPipeline
|
| 11 |
from depthcrafter.unet import DiffusersUNetSpatioTemporalConditionModelDepthCrafter
|
| 12 |
-
from depthcrafter.utils import vis_sequence_depth, save_video
|
| 13 |
-
|
| 14 |
-
dataset_res_dict = {
|
| 15 |
-
"sintel": [448, 1024],
|
| 16 |
-
"scannet": [640, 832],
|
| 17 |
-
"KITTI": [384, 1280],
|
| 18 |
-
"bonn": [512, 640],
|
| 19 |
-
"NYUv2": [448, 640],
|
| 20 |
-
}
|
| 21 |
|
| 22 |
|
| 23 |
class DepthCrafterDemo:
|
|
@@ -59,45 +50,6 @@ class DepthCrafterDemo:
|
|
| 59 |
print("Xformers is not enabled")
|
| 60 |
self.pipe.enable_attention_slicing()
|
| 61 |
|
| 62 |
-
@staticmethod
|
| 63 |
-
def read_video_frames(
|
| 64 |
-
video_path, process_length, target_fps, max_res, dataset="open"
|
| 65 |
-
):
|
| 66 |
-
if dataset == "open":
|
| 67 |
-
print("==> processing video: ", video_path)
|
| 68 |
-
vid = VideoReader(video_path, ctx=cpu(0))
|
| 69 |
-
print(
|
| 70 |
-
"==> original video shape: ", (len(vid), *vid.get_batch([0]).shape[1:])
|
| 71 |
-
)
|
| 72 |
-
original_height, original_width = vid.get_batch([0]).shape[1:3]
|
| 73 |
-
height = round(original_height / 64) * 64
|
| 74 |
-
width = round(original_width / 64) * 64
|
| 75 |
-
if max(height, width) > max_res:
|
| 76 |
-
scale = max_res / max(original_height, original_width)
|
| 77 |
-
height = round(original_height * scale / 64) * 64
|
| 78 |
-
width = round(original_width * scale / 64) * 64
|
| 79 |
-
else:
|
| 80 |
-
height = dataset_res_dict[dataset][0]
|
| 81 |
-
width = dataset_res_dict[dataset][1]
|
| 82 |
-
|
| 83 |
-
vid = VideoReader(video_path, ctx=cpu(0), width=width, height=height)
|
| 84 |
-
|
| 85 |
-
fps = vid.get_avg_fps() if target_fps == -1 else target_fps
|
| 86 |
-
stride = round(vid.get_avg_fps() / fps)
|
| 87 |
-
stride = max(stride, 1)
|
| 88 |
-
frames_idx = list(range(0, len(vid), stride))
|
| 89 |
-
print(
|
| 90 |
-
f"==> downsampled shape: {len(frames_idx), *vid.get_batch([0]).shape[1:]}, with stride: {stride}"
|
| 91 |
-
)
|
| 92 |
-
if process_length != -1 and process_length < len(frames_idx):
|
| 93 |
-
frames_idx = frames_idx[:process_length]
|
| 94 |
-
print(
|
| 95 |
-
f"==> final processing shape: {len(frames_idx), *vid.get_batch([0]).shape[1:]}"
|
| 96 |
-
)
|
| 97 |
-
frames = vid.get_batch(frames_idx).asnumpy().astype("float32") / 255.0
|
| 98 |
-
|
| 99 |
-
return frames, fps
|
| 100 |
-
|
| 101 |
def infer(
|
| 102 |
self,
|
| 103 |
video: str,
|
|
@@ -116,7 +68,7 @@ class DepthCrafterDemo:
|
|
| 116 |
):
|
| 117 |
set_seed(seed)
|
| 118 |
|
| 119 |
-
frames, target_fps =
|
| 120 |
video,
|
| 121 |
process_length,
|
| 122 |
target_fps,
|
|
|
|
| 3 |
import numpy as np
|
| 4 |
import torch
|
| 5 |
|
|
|
|
| 6 |
from diffusers.training_utils import set_seed
|
| 7 |
from fire import Fire
|
| 8 |
|
| 9 |
from depthcrafter.depth_crafter_ppl import DepthCrafterPipeline
|
| 10 |
from depthcrafter.unet import DiffusersUNetSpatioTemporalConditionModelDepthCrafter
|
| 11 |
+
from depthcrafter.utils import vis_sequence_depth, save_video, read_video_frames
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
|
| 14 |
class DepthCrafterDemo:
|
|
|
|
| 50 |
print("Xformers is not enabled")
|
| 51 |
self.pipe.enable_attention_slicing()
|
| 52 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
def infer(
|
| 54 |
self,
|
| 55 |
video: str,
|
|
|
|
| 68 |
):
|
| 69 |
set_seed(seed)
|
| 70 |
|
| 71 |
+
frames, target_fps = read_video_frames(
|
| 72 |
video,
|
| 73 |
process_length,
|
| 74 |
target_fps,
|