Spaces:
Sleeping
Sleeping
| import math | |
| import random | |
| import gradio as gr | |
| import os | |
| import json | |
| # Paths to the JSON files | |
| json_files = { | |
| "Kinetics700": "kinetics700_tune_.json", | |
| "STAR-benchmark": "starb_tune_.json", | |
| "FineDiving": "finediving_tune_.json" | |
| } | |
| VIDEO_NAME = 'video_name' | |
| QUESTION = 'question' | |
| LABEL = 'label' | |
| PREDICTION = 'prediction' | |
| left_side_columns = [VIDEO_NAME] | |
| right_side_columns = [QUESTION, LABEL, PREDICTION] | |
| batch_size = 8 | |
| target_size = (1024, 1024) | |
| def func(index, dataset): | |
| json_file = json_files[dataset] | |
| start_index = index * batch_size | |
| end_index = start_index + batch_size | |
| with open(json_file, 'r') as f: | |
| data = json.load(f) | |
| all_examples = data[start_index:end_index] | |
| values_lst = [] | |
| for example_idx, example in enumerate(all_examples): | |
| values = get_instance_values(example, dataset) | |
| values_lst += values | |
| return values_lst | |
| def get_instance_values(example, dataset_name): | |
| example[VIDEO_NAME] = os.path.abspath(os.path.join(dataset_name, example[VIDEO_NAME])) | |
| values = [] | |
| for k in left_side_columns + right_side_columns: | |
| value = example[k] | |
| values.append(value) | |
| return values | |
| demo = gr.Blocks() | |
| def get_col(example, dataset_name): | |
| instance_values = get_instance_values(example, dataset_name) | |
| with gr.Column(): | |
| inputs_left = [] | |
| assert len(left_side_columns) == len(instance_values[:len(left_side_columns)]) # excluding the video | |
| for key, value in zip(left_side_columns, instance_values[:len(left_side_columns)]): | |
| if key == VIDEO_NAME: | |
| if os.path.exists(value): # Check if the video file exists | |
| input_k = gr.Video(value=value) | |
| else: | |
| input_k = gr.Textbox(value=f"Video file not found: {value}", label=f"{key.capitalize()}") | |
| else: | |
| label = key.capitalize().replace("_", " ") | |
| input_k = gr.Textbox(value=value, label=f"{label}") | |
| inputs_left.append(input_k) | |
| with gr.Accordion("Click for details", open=False): | |
| text_inputs_right = [] | |
| assert len(right_side_columns) == len(instance_values[len(left_side_columns):]) | |
| for key, value in zip(right_side_columns, instance_values[len(left_side_columns):]): | |
| label = key.capitalize().replace("_", " ") | |
| if key == PREDICTION: | |
| text_input_k = gr.Textbox(value=value, label=f"{label}", lines=7) | |
| else: | |
| text_input_k = gr.Textbox(value=value, label=f"{label}") | |
| text_inputs_right.append(text_input_k) | |
| return inputs_left, text_inputs_right | |
| with demo: | |
| gr.Markdown("# Slide to iterate videos") | |
| with gr.Column(): | |
| dataset_dropdown = gr.Dropdown(choices=list(json_files.keys()), label="Select Dataset", value="Kinetics700") | |
| slider = gr.Slider(minimum=0, maximum=math.ceil(500 / batch_size), step=1, label='Page') # Assuming 500 samples per dataset | |
| with gr.Row(): | |
| index = slider.value | |
| dataset = dataset_dropdown.value | |
| start_index = 0 * batch_size | |
| end_index = start_index + batch_size | |
| with open(json_files[dataset], 'r') as f: | |
| data = json.load(f) | |
| all_examples = data[start_index:end_index] | |
| all_inputs_left_right = [] | |
| for example_idx, example in enumerate(all_examples): | |
| inputs_left, text_inputs_right = get_col(example, dataset) | |
| inputs_left_right = inputs_left + text_inputs_right | |
| all_inputs_left_right += inputs_left_right | |
| slider.change(func, inputs=[slider, dataset_dropdown], outputs=all_inputs_left_right) | |
| dataset_dropdown.change(func, inputs=[slider, dataset_dropdown], outputs=all_inputs_left_right) | |
| demo.launch() |