kirimaru/Bat_Identification

1"""2 3Step 1: Run Your Script4Save the above code in a file, for example, app.py. Then, run the script using Python:5 6python app.py7 8Step 2: Access the Gradio Interface9After running the script, you should see output in the terminal indicating that the Gradio interface is running. It will provide a local URL (usually http://localhost:7860) that you can open in your web browser to interact with your Gradio app.10 11Additional Features12Gradio supports various input and output types, including images, audio, and more. You can customize the interface further by adding more inputs and outputs, changing the layout, and adding examples.13 14For more advanced usage and features, you can refer to the Gradio documentation.15 16"""17 18import gradio as gr19import matplotlib.pyplot as plt20import pandas as pd21import numpy as np22 23import bat_detect.utils.detector_utils as du24import bat_detect.utils.audio_utils as au25import bat_detect.utils.plot_utils as viz26 27 28# setup the arguments29args = {}30args = du.get_default_bd_args()31args['detection_threshold'] = 0.332args['time_expansion_factor'] = 133args['model_path'] = 'models/Net2DFast_UK_same.pth.tar'34max_duration = 10.035 36# load the model37model, params = du.load_model(args['model_path'])38 39 40prediction_df = gr.Dataframe(41        headers=["species", "time", "detection_prob", "species_prob"],42        datatype=["str", "str", "str", "str"],43        row_count=1,44        col_count=(4, "fixed"),45        # max_height=300,46        elem_classes="qa-pairs",47        label='Predictions'48    )49 50prediction_css = """51.qa-pairs .table-wrap {52    min-height: 300px;53    max-height: 300px;54}55 56#visualisation-container {57  overflow-x: auto;58  width: 100%;59  height: 200px; /* Fixed height */60}61 62#visualisation-container img {63  height: 100%; /* fixed height */64  width: auto; /* flexible width */65  object-fit: contain; 66}67"""68 69examples = [['example_data/audio/20170701_213954-MYOMYS-LR_0_0.5.wav', 0.3],70            ['example_data/audio/20180530_213516-EPTSER-LR_0_0.5.wav', 0.3],71            ['example_data/audio/20180627_215323-RHIFER-LR_0_0.5.wav', 0.3],72            ['example_data/audio/Myotis daubentonii_A004034_PVTRQZEYAT.flac', 0.3],73            # ['example_data/audio/Eptesicus serotinus_A003974_RLNMCAZDEJ.flac', 0.3],74            # ['example_data/audio/Nyctalus noctula_A004093_DOEYCSPEZD.flac', 0.3],75]76 77 78def make_prediction(file_name=None, detection_threshold=0.3):79    80    if file_name is not None:81        audio_file = file_name82    else:83        return "You must provide an input audio file."84    85    if detection_threshold is not None and detection_threshold != '':86        args['detection_threshold'] = float(detection_threshold)87    88    # process the file to generate predictions89    results = du.process_file(audio_file, model, params, args, max_duration=max_duration)90    # results = du.process_file(audio_file, model, params, args, max_duration=False)91    92    anns = [ann for ann in results['pred_dict']['annotation']]93    clss = [aa['class'] for aa in anns]94    st_time = [aa['start_time'] for aa in anns]95    cls_prob = [aa['class_prob'] for aa in anns]96    det_prob = [aa['det_prob'] for aa in anns]97    data = {'species': clss, 'time': st_time, 'detection_prob': det_prob, 'species_prob': cls_prob}98    99    prediction_df = pd.DataFrame(data=data)100    im = generate_results_image(audio_file, anns)101    102    return [prediction_df, im]103 104 105def generate_results_image(audio_file, anns): 106    107    # load audio108    sampling_rate, audio = au.load_audio_file(audio_file, args['time_expansion_factor'], 109                           params['target_samp_rate'], params['scale_raw_audio'], max_duration=max_duration)110    duration = audio.shape[0] / sampling_rate111        112    # generate spec113    spec, spec_viz = au.generate_spectrogram(audio, sampling_rate, params, True, False)114 115    # create fig116    plt.close('all')117    # Adjust figsize calculation to control the width of the spectrogram118    fig_width = max(6, spec.shape[1] / 100)  # Minimum width of 6 inches, scales with spectrogram width119    fig_height = spec.shape[0] / 100 # Adjust the divisor for height scaling (higher = smaller)120    fig = plt.figure(1, figsize=(fig_width, fig_height), dpi=100, frameon=False)121 122    spec_duration = au.x_coords_to_time(spec.shape[1], sampling_rate, params['fft_win_length'], params['fft_overlap'])123    viz.create_box_image(spec, fig, anns, 0, spec_duration, spec_duration, params, spec.max()*1.1, False, True)124    plt.ylabel('Freq - kHz')125    plt.xlabel('Time - secs')126    plt.tight_layout()127    128    # convert fig to image129    fig.canvas.draw()130    data = np.frombuffer(fig.canvas.buffer_rgba(), dtype=np.uint8)131    w, h = fig.canvas.get_width_height()132    im = data.reshape((int(h), int(w), -1))133 134    return im135 136 137descr_txt = "Demo of Bat identification tools. " \138          "<br>It is based on two state-of-the-art models BatDetect2 and Bat-cli. " \139          "For the demo purposes, the input file is longer than 10 seconds, only the first 10 seconds will be processed." \140        #   "<br>Check out the two papers for more details [here](https://www.biorxiv.org/content/10.1101/2022.12.14.520490v1) and [here](https://ar5iv.labs.arxiv.org/html/2309.11218)."141 142Gradio_interface = gr.Interface(143    fn = make_prediction,144    inputs = [gr.Audio(sources=["upload"], type="filepath"), 145              gr.Dropdown([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9])],146    outputs = [prediction_df, gr.Image(interactive=False, label="Visualisation", elem_id="visualisation-container")],147    # theme = "huggingface",148    title = "Bat Identification Demo",149    description = descr_txt,150    examples = examples,151    allow_flagging = 'never',152    css=prediction_css153)154 155Gradio_interface.launch()156