progressbar

TrailBlazer/Pipeline/TextToVideoSDPipelineCall.py

@@ -1,3 +1,4 @@
+import tqdm
 import inspect
 from typing import Any, Callable, Dict, List, Optional, Union
 
@@ -62,6 +63,7 @@ def text_to_video_sd_pipeline_call(
     callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
     callback_steps: int = 1,
     cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+    progress = None,
 ):
     r"""
     The call function to the pipeline for generation.
@@ -251,7 +253,13 @@ def text_to_video_sd_pipeline_call(
     latents_at_steps = []
 
     with self.progress_bar(total=num_inference_steps) as progress_bar:
-        for i, t in enumerate(timesteps):
+
+        if type(progress)!=type(None):
+            timesteps = progress.tqdm(timesteps, desc="Processing")
+
+        i = 0
+        for t in timesteps:
+
             # expand the latents if we are doing classifier free guidance
             latent_model_input = (
                 torch.cat([latents] * 2) if do_classifier_free_guidance else latents
@@ -318,6 +326,7 @@ def text_to_video_sd_pipeline_call(
                 progress_bar.update()
                 if callback is not None and i % callback_steps == 0:
                     callback(i, t, latents)
+            i += 1
 
     if output_type == "latent":
         return TextToVideoSDPipelineOutput(frames=latents)

app.py

@@ -1,5 +1,6 @@
-import spaces
+#import spaces
 import sys
+import time
 import os
 import torch
 import gradio as gr
@@ -15,6 +16,7 @@ static = os.path.join(root, "static")
 from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler
 from diffusers.pipelines import TextToVideoSDPipeline
 from diffusers.utils import export_to_video
+
 from TrailBlazer.Misc import ConfigIO
 from TrailBlazer.Misc import Logger as log
 from TrailBlazer.Pipeline.TextToVideoSDPipelineCall import (
@@ -31,16 +33,14 @@ unet3d_condition_model_forward_copy = UNet3DConditionModel.forward
 UNet3DConditionModel.forward = unet3d_condition_model_forward
 
 
-from diffusers.utils import export_to_video
 
 model_id = "cerspense/zeroscope_v2_576w"
-model_path = model_id
+model_path = sys.argv[-1] + model_id
 pipe = DiffusionPipeline.from_pretrained(model_path, torch_dtype=torch.float16)
 pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
-#pipe.enable_model_cpu_offload()
 pipe.to('cuda')
 
-@spaces.GPU(duration=100)
+#@spaces.GPU(duration=120)
 def core(bundle):
     generator = torch.Generator().manual_seed(int(bundle["seed"]))
     result = pipe(
@@ -49,6 +49,7 @@ def core(bundle):
         width=512,
         generator=generator,
         num_inference_steps=40,
+        progress=gr.Progress(track_tqdm=True),
     )
     return result.frames
 
@@ -70,6 +71,7 @@ def gen_btn_fn(
     temporal_strengthen_scale,
     temporal_weaken_scale,
     rand_seed,
+    progress = gr.Progress(),
 ):
 
     bundle = {}
@@ -100,8 +102,6 @@ def gen_btn_fn(
     bboxes.pop()
     prompts.pop()
 
-
-
     for i in range(len(frames)):
         keyframe = {}
         keyframe["bbox_ratios"] = [float(v) for v in bboxes[i].split(",")]
@@ -208,10 +208,15 @@ with gr.Blocks(
     <li>Basic: The bounding box (bbox) is the tuple of four floats for the rectangular corners: left, top, right, bottom in the normalized ratio. The Word prompt indices is a list of 1-indexed numbers determining the prompt word.</li>
     <li>Advanced Options: We also offer some key parameters to adjust the synthesis result. Please see our paper for more information about the ablations.</li>
     </ul>
+
+    For your initial use, it is advisable to select one of the examples provided below and attempt to swap the subject first (e.g., cat -> lion). Subsequently, define the keyframe with the associated bbox/frame/prompt. Please note that our current work is based on the ZeroScope (cerspense/zeroscope_v2_576w) model. Using prompts that are commonly recognized in the ZeroScope model context is recommended.
     </p>
     """
-    gr.HTML(description)
 
+    gr.HTML(description)
+    dummy_note = gr.Textbox(
+        interactive=True, label="Note", visible=False
+    )
     with gr.Row():
         with gr.Column(scale=2):
             with gr.Row():
@@ -227,14 +232,16 @@ with gr.Blocks(
                         word_prompt_indices_tb = gr.Textbox(
                             interactive=True, label="Word prompt indices:"
                         )
-                        text = "Hint: Each keyframe ends with <strong>SEMICOLON</strong>, and <strong>COMMA</strong> for separating each value in the keyframe. The prompt field can be a single prompt without semicolon, or multiple prompts ended semicolon. One can use the SketchPadHelper tab to help to design the bboxes field."
+                        text = "<strong>Hint</strong>: Each keyframe ends with <strong>SEMICOLON</strong>, and <strong>COMMA</strong> for separating each value in the keyframe. The prompt field can be a single prompt without semicolon, or multiple prompts ended semicolon. One can use the SketchPadHelper tab to help to design the bboxes field."
                         gr.HTML(text)
                     with gr.Row():
                         clear_btn = gr.Button(value="Clear")
                         gen_btn = gr.Button(value="Generate")
 
                     with gr.Accordion("Advanced Options", open=False):
-                        text = "Hint: This default value should be sufficient for most tasks. However, it's important to note that our approach is currently implemented on ZeroScope, and its performance may be influenced by the model's characteristics. We plan to conduct experiments on different models in the future."
+                        text = "<strong>Hint</strong>: This default value should be sufficient for most tasks. However, it's important to note that our approach is currently implemented on ZeroScope, and its performance may be influenced by the model's characteristics. We plan to conduct experiments on different models in the future."
+                        gr.HTML(text)
+                        text = "<strong>Hint</strong>: When the #Spatial edits and #Temporal edits sliders are 0, it means the experiment will run without TrailBlazer but just simply a T2V generation through ZeroScope."
                         gr.HTML(text)
                         with gr.Row():
                             trailing_length = gr.Slider(
@@ -321,7 +328,7 @@ with gr.Blocks(
                             out_board_cb, inputs=[user_board], outputs=[out_board]
                         )
                     with gr.Row():
-                        text = "Hint: Utilize a black pen with the Draw Button to create a ``rough'' bbox. When you press the green ``Save Changes'' Button, the app calculates the minimum and maximum boundaries. Each ``Layer'', located at the bottom left of the pad, corresponds to one bounding box. Copy the returned value to the bbox textfield in the main tab."
+                        text = "<strong>Hint</strong>: Utilize a black pen with the Draw Button to create a ``rough'' bbox. When you press the green ``Save Changes'' Button, the app calculates the minimum and maximum boundaries. Each ``Layer'', located at the bottom left of the pad, corresponds to one bounding box. Copy the returned value to the bbox textfield in the main tab."
                         gr.HTML(text)
                     with gr.Row():
                         out_label = gr.Label(label="Converted bboxes string")
@@ -331,7 +338,7 @@ with gr.Blocks(
 
         with gr.Column(scale=1):
             gr.HTML(
-                '<span style="font-size: 20px; font-weight: bold">Generated Images</span>'
+                '<span style="font-size: 20px; font-weight: bold">Generated Video</span>'
             )
             with gr.Row():
                 out_gen_1 = gr.Video(visible=True, show_label=False)
@@ -340,11 +347,12 @@ with gr.Blocks(
         gr.Examples(
             examples=[
                 [
-                    "A clown fish swimming in a coral reef",
+                    "A clownfish swimming in a coral reef",
                     "0.5,0.35,1.0,0.65; 0.0,0.35,0.5,0.65;",
                     "0; 24;",
-                    "1,2,3",
+                    "1, 2",
                     "123451232531",
+                    "It generates clownfish at right, then move to left",
                     "assets/gradio/fish-RL.mp4",
                 ],
                 [
@@ -352,8 +360,9 @@ with gr.Blocks(
                     "0.0,0.35,0.4,0.65; 0.6,0.35,1.0,0.65; 0.0,0.35,0.4,0.65;"
                     "0.6,0.35,1.0,0.65; 0.0,0.35,0.4,0.65;",
                     "0; 6; 12; 18; 24;",
-                    "1,2",
+                    "1, 2",
                     "123451232530",
+                    "The cat will run Left/Right/Left/Right",
                     "assets/gradio/cat-LRLR.mp4",
                 ],
                 [
@@ -362,6 +371,7 @@ with gr.Blocks(
                     "0; 24;",
                     "1, 2",
                     "0",
+                    "The fish moves from top left to bottom right, from far to near.",
                     "assets/gradio/fish-TL2BR.mp4"
                 ],
                 [
@@ -370,6 +380,7 @@ with gr.Blocks(
                     "0; 24;",
                     "1, 2",
                     "0",
+                    "Same with the above but now the prompt associates with tiger",
                     "assets/gradio/tiger-TL2BR.mp4"
                 ],
                 [
@@ -378,10 +389,11 @@ with gr.Blocks(
                     "0; 24;",
                     "1,2,3",
                     "123451232531",
+                    "The subject will deformed from cat to dog.",
                     "assets/gradio/Cat2Dog.mp4",
                 ],
             ],
-            inputs=[text_prompt_tb, bboxes_tb, frame_tb, word_prompt_indices_tb, rand_seed,out_gen_1],
+            inputs=[text_prompt_tb, bboxes_tb, frame_tb, word_prompt_indices_tb, rand_seed, dummy_note, out_gen_1],
             outputs=None,
             fn=None,
             cache_examples=False,
@@ -411,8 +423,7 @@ with gr.Blocks(
             rand_seed,
         ],
         outputs=[out_gen_1],
-        queue=False,
     )
 
-main.queue(max_size=40, api_open=False)
-main.launch(max_threads=400)
+main.launch()
+# main.launch(max_threads=400)