update app.py

app.py

@@ -23,8 +23,6 @@ model.eval()
 avg_pool = torch.nn.AdaptiveAvgPool2d((1, 1))
 
 def predict(input_file, downsample_size):
-    downsample_size = int(downsample_size)
-
     base_directory = os.getcwd()
     selected_directory = os.path.join(base_directory, "selected_images")
     if os.path.isdir(selected_directory):
@@ -53,8 +51,8 @@ def predict(input_file, downsample_size):
     img_vecs = np.asarray(img_vecs)
     print("images encoded")
     rv_indices, _ = furthest_neighbours(
-        img_vecs,
-        downsample_size,
+        x=img_vecs,
+        downsample_size=downsample_size,
         seed=0)
     indices = np.zeros((img_vecs.shape[0],))
     indices[np.asarray(rv_indices)] = 1
@@ -88,7 +86,7 @@ demo = gr.Interface(
     title="Frame selection by visual difference",
     fn=predict,
     inputs=[gr.components.Video(label="Upload Video File"),
-            gr.components.Number(label="Downsample size")],
+            gr.components.Number(label="Downsample size", precision=0)],
     outputs=gr.components.File(label="Zip"),
 )
 

sampling_util.py

@@ -3,14 +3,14 @@ import numpy as np
 from fastdist2 import euclidean_vector_to_matrix_distance
 
 
-def furthest_neighbours(x, downsampled_size, seed):
+def furthest_neighbours(x, downsample_size, seed):
     x = x.astype(np.float32)
     np.random.seed(seed)
     length = x.shape[0]
     img_vecs_dims = x.shape[-1]
 
-    rv_indices = [np.random.randint(low=0, high=downsampled_size - 1, size=1)[0]]
-    selected_points = np.zeros((downsampled_size, img_vecs_dims), np.float32)
+    rv_indices = [np.random.randint(low=0, high=downsample_size - 1, size=1)[0]]
+    selected_points = np.zeros((downsample_size, img_vecs_dims), np.float32)
     selected_points[0, :] = x[rv_indices[0], :]
 
     distance_for_selected_min = np.ones((length,)) * 1e15
@@ -18,7 +18,7 @@ def furthest_neighbours(x, downsampled_size, seed):
     inactive_points = np.zeros(length, dtype=bool)
     inactive_points[rv_indices[0]] = True
 
-    for i in (range(downsampled_size - 1)):
+    for i in (range(downsample_size - 1)):
         distance_for_selected = euclidean_vector_to_matrix_distance(selected_points[i, :], x)
         distance_for_selected_min = np.minimum(distance_for_selected_min, distance_for_selected)
         furthest_point_idx = np.argmax(np.ma.array(distance_for_selected_min, mask=inactive_points))