File size: 13,441 Bytes
0659b98
7c35e92
 
0659b98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
72a0cd6
7c35e92
6825e25
7c35e92
366a2ee
7c35e92
 
 
 
72a0cd6
7c35e92
 
95687fc
7c35e92
 
 
 
 
 
 
0659b98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c8296fc
7c35e92
 
0659b98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c8296fc
 
 
 
 
9e96e5e
7c35e92
0659b98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9e96e5e
0659b98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9e96e5e
0659b98
 
 
 
 
 
 
 
 
 
 
c8296fc
0659b98
7c35e92
0659b98
 
 
 
 
7c35e92
0659b98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c8296fc
0659b98
7c35e92
0659b98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
import os
import torch
import cv2
import subprocess
from datetime import datetime
from pathlib import Path
import gradio as gr
import numpy as np
from huggingface_hub import hf_hub_download, snapshot_download
# -----------------------------
# Setup paths and env
# -----------------------------
HF_HOME = "/app/hf_cache"
os.environ["HF_HOME"] = HF_HOME
os.environ["TRANSFORMERS_CACHE"] = HF_HOME
os.makedirs(HF_HOME, exist_ok=True)


PRETRAINED_DIR = "/app/pretrained"
os.makedirs(PRETRAINED_DIR, exist_ok=True)

# -----------------------------
# Step 1: Optional Model Download
# -----------------------------
def download_models():
    expected_model = os.path.join(PRETRAINED_DIR, "RAFT/raft-things.pth")
    if not Path(expected_model).exists():
        print("⚙️ Downloading pretrained models...")
        try:
            subprocess.check_call(["bash", "download/download_models.sh"])
            print("✅ Models downloaded.")
        except subprocess.CalledProcessError as e:
            print(f"Model download failed: {e}")
    else:
        print("✅ Pretrained models already exist.")


def visualize_depth_npy_as_video(npy_file, fps):
    # Load .npy file
    depth_np = np.load(npy_file)  # Shape: [T, 1, H, W]
    tensor = torch.from_numpy(depth_np)
    T, _, H, W = tensor.shape

    # Prepare video writer
    video_path = "depth_video_preview.mp4"
    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    out = cv2.VideoWriter(video_path, fourcc, fps, (W, H))  # 10 FPS

    for i in range(T):
        frame = tensor[i, 0].numpy()
        norm = (frame - frame.min()) / (frame.max() - frame.min() + 1e-8)
        frame_uint8 = (norm * 255).astype(np.uint8)
        colored = cv2.applyColorMap(frame_uint8, cv2.COLORMAP_INFERNO)
        out.write(colored)

    out.release()
    return video_path
# -----------------------------
# Step 1: Get Anchor Video
# -----------------------------
def get_anchor_video(video_path, fps, num_frames, target_pose, mode,
                       radius_scale, near_far_estimated,
                       sampler_name, diffusion_guidance_scale, diffusion_inference_steps,
                       prompt, negative_prompt, refine_prompt,
                       depth_inference_steps, depth_guidance_scale,
                       window_size, overlap, max_res, sample_size,
                       seed_input, height, width, aspect_ratio_inputs,
                       init_dx, init_dy, init_dz):

    temp_input_path = "/app/temp_input.mp4"
    output_dir = "/app/output_anchor"
    video_output_path = f"{output_dir}/masked_videos/output.mp4"
    captions_text_file = f"{output_dir}/captions/output.txt"
    depth_file = f"{output_dir}/depth/output.npy"


    if video_path:
        os.system(f"cp '{video_path}' {temp_input_path}")

    try:
        theta, phi, r, x, y = target_pose.strip().split()
    except ValueError:
        return f"Invalid target pose format. Use: θ φ r x y", None, None
    logs =  f"Running inference with target pose: θ={theta}, φ={phi}, r={r}, x={x}, y={y}\n"
    w, h = aspect_ratio_inputs.strip().split(",")
    h_s, w_s = sample_size.strip().split(",")
    
    command = [
        "python", "/app/inference/v2v_data/inference.py",
        "--video_path", temp_input_path,
        "--stride", "1",
        "--out_dir", output_dir,
        "--radius_scale", str(radius_scale),
        "--camera", "target",
        "--mask",
        "--target_pose", theta, phi, r, x, y,
        "--video_length", str(num_frames),
        "--save_name", "output",
        "--mode", mode,
        "--fps", str(fps),
        "--depth_inference_steps", str(depth_inference_steps),
        "--depth_guidance_scale", str(depth_guidance_scale),
        "--near_far_estimated", str(near_far_estimated),
        "--sampler_name", sampler_name,
        "--diffusion_guidance_scale", str(diffusion_guidance_scale),
        "--diffusion_inference_steps", str(diffusion_inference_steps),
        "--prompt", prompt if prompt else "",
        "--negative_prompt", negative_prompt,
        "--refine_prompt", refine_prompt,
        "--window_size", str(window_size),
        "--overlap", str(overlap),
        "--max_res", str(max_res),
        "--sample_size", h_s.strip(), w_s.strip(),
        "--seed", str(seed_input),
        "--height", str(height),
        "--width", str(width),
        "--target_aspect_ratio", w.strip(), h.strip(),
        "--init_dx", str(init_dx),
        "--init_dy", str(init_dy),
        "--init_dz", str(init_dz),
  
    ]   

    try:
        result = subprocess.run(command, capture_output=True, text=True, check=True)
        logs += result.stdout
    except subprocess.CalledProcessError as e:
        logs += f"Inference failed:\n{e.stderr}{e.stdout}"
        return None, logs
    
    caption_text = ""
    if os.path.exists(captions_text_file):
        with open(captions_text_file, "r") as f:
            caption_text = f.read()
    depth_video_path = visualize_depth_npy_as_video(depth_file, fps)
    return str(video_output_path), logs, caption_text, depth_video_path
# -----------------------------
# Step 2: Run Inference
# -----------------------------
def inference(
    fps, num_frames, controlnet_weights, controlnet_guidance_start,
    controlnet_guidance_end, guidance_scale, num_inference_steps, dtype,
    seed, height, width, downscale_coef, vae_channels,
    controlnet_input_channels, controlnet_transformer_num_layers
):
    model_path = "/app/pretrained/CogVideoX-5b-I2V"
    ckpt_path = "/app/out/EPiC_pretrained/checkpoint-500.pt"
    video_root_dir = "/app/output_anchor"
    out_dir = "/app/output"

    command = [
        "python", "/app/inference/cli_demo_camera_i2v_pcd.py",
        "--video_root_dir", video_root_dir,
        "--base_model_path", model_path,
        "--controlnet_model_path", ckpt_path,
        "--output_path", out_dir,
        "--controlnet_weights", str(controlnet_weights),
        "--controlnet_guidance_start", str(controlnet_guidance_start),
        "--controlnet_guidance_end", str(controlnet_guidance_end),
        "--guidance_scale", str(guidance_scale),
        "--num_inference_steps", str(num_inference_steps),
        "--dtype", dtype,
        "--seed", str(seed),
        "--height", str(height),
        "--width", str(width),
        "--num_frames", str(num_frames),
        "--fps", str(fps),
        "--downscale_coef", str(downscale_coef),
        "--vae_channels", str(vae_channels),
        "--controlnet_input_channels", str(controlnet_input_channels),
        "--controlnet_transformer_num_layers", str(controlnet_transformer_num_layers),

    ]

    try:
        result = subprocess.run(command, capture_output=True, text=True, check=True)
        logs = result.stdout
    except subprocess.CalledProcessError as e:
        logs = f"❌ Step 2 Inference Failed:\nSTDERR:\n{e.stderr}\nSTDOUT:\n{e.stdout}"
        return None, logs
    video_output = f"{out_dir}/00000_{seed}_out.mp4"
    return video_output if os.path.exists(video_output) else None, logs

# -----------------------------
# UI
# -----------------------------
demo = gr.Blocks()

with demo:
    gr.Markdown("## 🎬 EPiC: Cinematic Camera Control")

    with gr.Tabs():
        with gr.TabItem("Step 1: Camera Anchor"):
            with gr.Row():
                with gr.Column():
                    with gr.Row():
                        near_far_estimated = gr.Checkbox(label="Near Far Estimation", value=True) 
                        pose_input = gr.Textbox(label="Target Pose (θ φ r x y)", placeholder="e.g., 0 30 -0.6 0 0")
                        fps_input = gr.Number(value=24, label="FPS")
                        aspect_ratio_inputs=gr.Textbox(value= "3,4",label="Target Aspect Ratio (e.g., 2,3)")

                        init_dx = gr.Number(value=0.0, label="Start Camera Offset X")
                        init_dy = gr.Number(value=0.0, label="Start Camera Offset Y")
                        init_dz = gr.Number(value=0.0, label="Start Camera Offset Z")

                        num_frames_input = gr.Number(value=49, label="Number of Frames")
                        radius_input = gr.Number(value = 1.0, label="Radius Scale")
                        mode_input = gr.Dropdown(choices=["gradual"], value="gradual", label="Camera Mode")
                        sampler_input = gr.Dropdown(choices=["Euler", "Euler A", "DPM++", "PNDM", "DDIM_Cog", "DDIM_Origin"], value="DDIM_Origin", label="Sampler")
                        diff_guidance_input = gr.Number(value=6.0, label="Diffusion Guidance")
                        diff_steps_input = gr.Number(value=50, label="Diffusion Steps")
                        depth_steps_input = gr.Number(value=5, label="Depth Steps")
                        depth_guidance_input = gr.Number(value=1.0, label="Depth Guidance")
                        window_input = gr.Number(value=64, label="Window Size")    
                        overlap_input = gr.Number(value=25, label="Overlap")
                        maxres_input = gr.Number(value=720, label="Max Resolution")
                        sample_size = gr.Textbox(label="Sample Size (height, width)", placeholder="e.g., 384, 672", value="384, 672")
                        seed_input = gr.Number(value=43, label="Seed")
                        height = gr.Number(value=480, label="Height")
                        width = gr.Number(value=720, label="Width")
                        prompt_input = gr.Textbox(label="Prompt")
                        neg_prompt_input = gr.Textbox(label="Negative Prompt", value="The video is not of a high quality, it has a low resolution. Watermark present in each frame. The background is solid. Strange body and strange trajectory.")
                        refine_prompt_input = gr.Textbox(label="Refine Prompt", value=" The video is of high quality, and the view is very clear. ")
                with gr.Column():
                    video_input = gr.Video(label="Upload Video (MP4)")
                    step1_button = gr.Button("▶️ Run Step 1")
                    step1_video = gr.Video(label="[Step 1] Masked Video")
                    step1_captions = gr.Textbox(label="[Step 1] Captions", lines=4)
                    step1_logs = gr.Textbox(label="[Step 1] Logs")
                    step1_depth = gr.Video(label="[Step 1] Depth Video", visible=False)  # Hidden by default

        with gr.TabItem("Step 2: CogVideoX Refinement"):
            with gr.Row():
                with gr.Column():
                    with gr.Row():
                  
                        controlnet_weights_input = gr.Number(value=0.5, label="ControlNet Weights")
                        controlnet_guidance_start_input = gr.Number(value=0.0, label="Guidance Start")
                        controlnet_guidance_end_input = gr.Number(value=0.5, label="Guidance End")
                        guidance_scale_input = gr.Number(value=6.0, label="Guidance Scale")
                        inference_steps_input = gr.Number(value=50, label="Num Inference Steps")
                        dtype_input = gr.Dropdown(choices=["float16", "bfloat16"], value="bfloat16", label="Compute Dtype")
                        seed_input2 = gr.Number(value=42, label="Seed")
                        height_input = gr.Number(value=480, label="Height")
                        width_input = gr.Number(value=720, label="Width")
                        num_frames_input2 = gr.Number(value=49, label="Num Frames")
                        fps_input2 = gr.Number(value=24, label="FPS")
                        downscale_coef_input = gr.Number(value=8, label="Downscale Coef")
                        vae_channels_input = gr.Number(value=16, label="VAE Channels")
                        controlnet_input_channels_input = gr.Number(value=6, label="ControlNet Input Channels")
                        controlnet_layers_input = gr.Number(value=8, label="ControlNet Transformer Layers")
                with gr.Column():
                    step2_video = gr.Video(label="[Step 2] Final Refined Video")
                    step2_button = gr.Button("▶️ Run Step 2")
                    step2_logs = gr.Textbox(label="[Step 2] Logs")


    step1_button.click(
        get_anchor_video,
        inputs=[
            video_input, fps_input, num_frames_input, pose_input, mode_input,
            radius_input, near_far_estimated,
            sampler_input, diff_guidance_input, diff_steps_input,
            prompt_input, neg_prompt_input, refine_prompt_input,
            depth_steps_input, depth_guidance_input,
            window_input, overlap_input, maxres_input, sample_size,
            seed_input, height, width, aspect_ratio_inputs,
            init_dx, init_dy, init_dz
        ],
        outputs=[step1_video, step1_logs, step1_captions, step1_depth]  # ← updated here
    )

    step2_button.click(
        inference,
        inputs=[
            fps_input2, num_frames_input2,
            controlnet_weights_input, controlnet_guidance_start_input,
            controlnet_guidance_end_input, guidance_scale_input,
            inference_steps_input, dtype_input, seed_input2,
            height_input, width_input, downscale_coef_input,
            vae_channels_input, controlnet_input_channels_input,
            controlnet_layers_input
        ],
        outputs=[step2_video, step2_logs]
    )

if __name__ == "__main__":
    download_models()
    demo.launch(server_name="0.0.0.0", server_port=7860)