Upload V1.py

V1.py

@@ -7,7 +7,6 @@ from torch import nn
 from torch.nn import functional as F
 import matplotlib.pyplot as plt
 import os
-import pandas as pd
 import imageio
 from torch.cuda.amp import autocast as autocast
 
@@ -744,166 +743,3 @@ def circular_hist(ax, x, bins=16, density=True, offset=0, gaps=True):
         ax.set_yticks([])
 
     return n, bins, patches
-
-
-def show_trained_model(file_name="/home/2TSSD/experiment/FFMEDNN/Sintel_fixv1_10.62_ckpt.pth.tar"):
-    import utils.torch_utils as utils
-    from model.fle_version_2_3.FFV1MT_MS import FFV1DNN
-    model = FFV1DNN(num_scales=8,
-                    num_cells=256,
-                    upsample_factor=8,
-                    feature_channels=256,
-                    scale_factor=16,
-                    num_layers=6)
-    # model = utils.restore_model(model, file_name)
-    model = model.ffv1
-    t_point = 100
-    s_point = 100
-    t_kz = 6
-    filenames = []
-    x = np.arange(0, 6) * 40
-    x = np.repeat(x[None], axis=0, repeats=256)
-    temporal = model.temporal_pooling.data.cpu().squeeze().numpy()
-    mean = np.mean(temporal, axis=0)
-    plt.figure(figsize=(10, 10))
-    plt.subplot(2, 1, 1)
-    for idx in range(0, 256):
-        plt.plot(x[idx], temporal[idx])
-    plt.subplot(2, 1, 2)
-    plt.plot(x[0], mean, label="mean")
-
-    plt.xlabel("times (ms)")
-    plt.ylabel("temporal pooling weight")
-    plt.legend()
-    plt.grid(True)
-    plt.show()
-    neural_representation = model._get_v1_order()
-
-    fs = np.array([ne["fs"] for ne in neural_representation])
-    ft = np.array([ne["ft"] for ne in neural_representation])
-
-    ax1 = plt.subplot(131, projection='polar')
-    theta_list = []
-    v_list = []
-    energy_list = []
-    for index in range(len(neural_representation)):
-        v = neural_representation[index]["speed"]
-        theta = neural_representation[index]["theta"]
-        theta_list.append(theta)
-        v_list.append(v)
-
-    v_list, theta_list = np.array(v_list), np.array(theta_list)
-    x, y = pol2cart(v_list, theta_list)
-    plt.scatter(theta_list, v_list, c=v_list, cmap="rainbow", s=(v_list + 20), alpha=0.8)
-    plt.axis('on')
-    # plt.colorbar()
-    plt.grid(True)
-    # plt.subplot(132, projection="polar")
-    # plt.scatter(theta_list, np.ones_like(theta_list))
-    plt.subplot(132, projection='polar')
-    plt.scatter(theta_list, np.ones_like(v_list))
-    lst = []
-    for scale in range(8):
-        lst += ["scale %d" % scale] * 32
-    data = {"Spatial Frequency": fs, 'Temporal Frequency': ft, "Class": lst}
-    df = pd.DataFrame(data=data)
-    ax = plt.subplot(133, projection='polar')
-    # theta_list = theta_list[v_list > (ft * v_list.mean())]
-    print(len(theta_list))
-    bins_number = 8  # the [0, 360) interval will be subdivided into this
-    # number of equal bins
-    zone = np.pi / 8
-    theta_list[theta_list < (-np.pi + zone)] = theta_list[theta_list < (-np.pi + zone)] + np.pi * 2
-    bins = np.linspace(-np.pi + zone, np.pi + zone, bins_number + 1)
-    n, _, _ = plt.hist(theta_list, bins, edgecolor="black")
-    # ax.set_theta_offset(-np.pi / 8 - np.pi)
-    ax.set_yticklabels([])
-    plt.grid(True)
-    import seaborn as sns
-    sns.jointplot(data=df, x="Spatial Frequency", y="Temporal Frequency", hue="Class", xlim=[0, 0.3], ylim=[0, 0.3])
-    plt.grid(True)
-    g = sns.jointplot(data=df, x="Spatial Frequency", y="Temporal Frequency", xlim=[0, 0.25], ylim=[0, 0.25])
-    # g.plot_joint(sns.kdeplot, color="r", zorder=0, levels=6)
-
-    plt.grid(True)
-    plt.show()
-
-    # show spatial frequency preference and temporal frequency preference.
-
-    x = np.linspace(0, t_kz, t_point)
-    index = 0
-    for scale in range(len(model.spatial_filter)):
-        t_sin, t_cos = model.temporal_filter[scale].demo_temporal_filter(points=t_point)
-        gb_sin_b, gb_cos_b = model.spatial_filter[scale].demo_gabor_filters(points=s_point)
-        for i in range(gb_sin_b.size(0)):
-            plt.figure(figsize=(14, 9), dpi=80)
-            plt.subplot(2, 3, 1)
-            curve = gb_sin_b[i].squeeze().detach().numpy()
-            plt.imshow(curve)
-            plt.title("Gabor Sin")
-            plt.subplot(2, 3, 2)
-            curve = gb_cos_b[i].squeeze().detach().numpy()
-            plt.imshow(curve)
-            plt.title("Gabor Cos")
-
-            plt.subplot(2, 3, 3)
-            curve = t_sin[i].squeeze().detach().numpy()
-            plt.plot(x, curve, label='sin')
-            plt.title("Temporal Sin")
-
-            curve = t_cos[i].squeeze().detach().numpy()
-            plt.plot(x, curve, label='cos')
-            plt.xlabel('Time (s)')
-            plt.ylabel('Response to pulse at t=0')
-            plt.legend()
-            plt.title("Temporal filter")
-
-            gb_sin = gb_sin_b[i].squeeze().detach()[5, :]
-            gb_cos = gb_cos_b[i].squeeze().detach()[5, :]
-
-            a = np.outer(t_cos[i].detach(), gb_sin)
-            b = np.outer(t_sin[i].detach(), gb_cos)
-            g_o = a + b
-
-            a = np.outer(t_sin[i].detach(), gb_sin)
-            b = np.outer(t_cos[i].detach(), gb_cos)
-            g_e = a - b
-            energy_component = g_o ** 2 + g_e ** 2
-
-            plt.subplot(2, 3, 4)
-            curve = g_o
-            plt.imshow(curve, cmap="gray")
-            plt.title("Spatial Temporal even")
-            plt.subplot(2, 3, 5)
-            curve = g_e
-            plt.imshow(curve, cmap="gray")
-            plt.title("Spatial Temporal odd")
-
-            plt.subplot(2, 3, 6)
-            curve = energy_component
-            plt.imshow(curve, cmap="gray")
-            plt.title("energy")
-            plt.savefig('filter_%d.png' % (index))
-            filenames.append('filter_%d.png' % (index))
-            index += 1
-            # plt.show()
-
-    # build gif
-    with imageio.get_writer('filters_orientation.gif', mode='I') as writer:
-        for filename in filenames:
-            image = imageio.imread(filename)
-            writer.append_data(image)
-
-    # Remove files
-    for filename in set(filenames):
-        os.remove(filename)
-
-
-if __name__ == "__main__":
-    show_trained_model()
-    # V1.demo()
-    # draw_polar()
-    # # V1.demo()
-    # # draw_polar()
-    show_trained_model()
-    # te_spatial_temporal()