plot_3x3_figure.py

import matplotlib.pyplot as plt
import soundfile as sf
import librosa

plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = 'Ubuntu'
plt.rcParams['font.monospace'] = 'Ubuntu Mono'
plt.rcParams['font.size'] = 10
plt.rcParams['axes.labelsize'] = 10   # Time and Hz, i.e. labels
plt.rcParams['axes.labelweight'] = 'bold'
plt.rcParams['xtick.labelsize'] = 15  # Time(x) tick values
plt.rcParams['ytick.labelsize'] = 15  # Hz(y) tick values
plt.rcParams['legend.fontsize'] = 17
plt.rcParams['figure.titlesize'] = 10
plt.rcParams['axes.titlesize'] = 14   # Title font

#---------------------------------------------------
# Plot amplitudes

fig, axarr = plt.subplots(3, 3, figsize=(17, 8))
fig.suptitle("15 sec & 3 sec & 1.5 sec", fontsize=16)

length = 5 # sec

#----------------------------------------------------
# Reference
ref, sr = sf.read("adele/test/11 Someone Like You.wav", 44100*15, )

axarr[0,0].plot(ref[:44100*length], 'b')
axarr[0,1].plot(ref[:44100*length//5], 'b')
axarr[0,2].plot(ref[:44100*length//10], 'b')
axarr[0,2].legend(['Reference'], loc=1)

#----------------------------------------------------
# Damaged
y, sr = librosa.core.load("adele/test/4410/11 Someone Like You.wav", sr=None)


axarr[1,0].plot(y[:4410*length], 'r')
axarr[1,1].plot(y[:4410*length//5], 'r')
axarr[1,2].plot(y[:4410*length//10], 'r')
axarr[1,2].legend(['Damaged'], loc=1)

#----------------------------------------------------
# Decoded
decoded, sr = librosa.core.load("adele/test/decoded/11 Someone Like You.mp3", sr=None)

axarr[2,0].plot(decoded[:44100*length], 'g')
axarr[2,1].plot(decoded[:44100*length//5], 'g')
axarr[2,2].plot(decoded[:44100*length//10], 'g')
axarr[2,2].legend(['Decoded'], loc=1)

#----------------------------------------------------
# Fine-tune figure; hide x ticks for top plots and y ticks for right plots
plt.setp([a.get_xticklabels() for a in axarr[0, :]], visible=False)
plt.setp([a.get_yticklabels() for a in axarr[:, 1]], visible=False)

# Tight layout often produces nice results
# but requires the title to be spaced accordingly
fig.tight_layout()
fig.subplots_adjust(top=0.92)

def plot_mean_video_wise_poster(total_data: ndarray, data_name:str, x_labels):
    '''
    Reference:
        - https://matplotlib.org/3.1.1/gallery/subplots_axes_and_figures/subplots_demo.html
    '''
    plt.rcParams['axes.labelsize'] = 20  # Label font size
    nrows = 6; ncols = 3
    fig, axarr = plt.subplots(
            nrows=nrows, ncols=ncols, figsize=(18, 18),
            sharex=True, sharey=True,
        )
    fig.suptitle(f'Unravel {data_name} facial expression embedding, \ncolumn-wise means before preprocess', fontsize=26)
    for video_id, data in zip(range(1, 18), total_data):
        #---------------------------------------------------
        # Plot amplitudes
        i = video_id - 1
        ax = axarr[i // ncols, i % ncols]
        ax.plot(
            range(len(data)),
            data
        )
        plt.ylim(ymax=1)
        ax.set_xticks(range(len(x_labels)))
        ax.set_xticklabels(x_labels)
        # Rotate the tick labels and set their alignment.
        plt.setp(ax.get_xticklabels(), rotation=45, ha="right",
             rotation_mode="anchor")

        ax.set_title(f'Video {video_id}')
        ax.set(xlabel='Class', ylabel='Mean', )
        ax.label_outer()
    plt.setp(axarr[-1, -1].get_xticklabels(), rotation=45, ha="right",
             rotation_mode="anchor")
       
    # Tight layout often produces nice results
    # but requires the title to be spaced accordingly
    fig.tight_layout()
    fig.subplots_adjust(top=0.92)
    plt.savefig(f'plots/data-analysis/poster-embedding={data_name}-columnwise_mean.png', bbox_inches = "tight")

classes=['Anger', 'Disgust', 'Fear', 'Happiness','Sadness','Surprise','Neutrality']
plot_mean_video_wise_poster(means, "Affectnet", x_labels=classes)

def plot_predictions_batch(images: List[ndarray], predictions: List[str], gts: List[str], fig_title: str):
    n_image_per_plot = 42
    n_image = len(images)
    for i in range(math.ceil(n_image / n_image_per_plot)):
        start_i = i * n_image_per_plot
        end_i = start_i + n_image_per_plot
        plot_predictions(
            images[start_i: end_i],
            predictions[start_i: end_i],
            gts[start_i: end_i],
            fig_title=f'{fig_title}_{i}'
        )

def plot_predictions(images: List[ndarray], predictions: List[str], gts: List[str], fig_title: str):
    n_image = len(images)
    n_col = 6
    n_row = math.ceil(n_image / n_col)
    fig, axes = plt.subplots(n_row, n_col)
    for i, (y, x) in zip(range(n_image), itertools.product(range(n_row), range(n_col))):
        ax = axes[y, x]
        ax.axis('off')
        ax.imshow(images[i])
        pred, gt = predictions[i], gts[i]
        if pred == gt:
            ax.set_title(gt, fontdict={'color':'green'}, fontsize=11)
        else:
            ax.set_title(f'gt: {gt}\npr: {pred}', fontdict={'color':'red'}, fontsize=11)

    fig.set_size_inches(n_col * 2, n_row * 0.9) 
    fig.suptitle(fig_title, fontsize=16)
    fig.savefig(f'plots/{fig_title}.jpg')
    
plot_predictions_batch(val_input_images, result_validation, val_labels, fig_title='pretrained=synth-val_set')