Modification to the eigen smoothing

pytorch_grad_cam/utils/svd_on_activations.py

@@ -1,19 +1,29 @@
-import numpy as np
 
 
-def get_2d_projection(activation_batch):
- # TBD: use pytorch batch svd implementation
- activation_batch[np.isnan(activation_batch)] = 0
- projections = []
- for activations in activation_batch:
- reshaped_activations = (activations).reshape(
- activations.shape[0], -1).transpose()
- # Centering before the SVD seems to be important here,
- # Otherwise the image returned is negative
- reshaped_activations = reshaped_activations - \
- reshaped_activations.mean(axis=0)
- U, S, VT = np.linalg.svd(reshaped_activations, full_matrices=True)
- projection = reshaped_activations @ VT[0, :]
- projection = projection.reshape(activations.shape[1:])
- projections.append(projection)
- return np.float32(projections)
+def get_2d_projection( batch_activations): 
+
+ batch_activations = torch.from_numpy( batch_activations)
+
+ b, c, h, w = batch_activations.shape
+
+ #x = rearrange(batch_activations, "b c h w -> b (h w) c")
+ x = batch_activations.reshape( b, c, h * w).permute( 0, 2, 1)
+
+ x_mean = x.mean(1, keepdim=True)
+
+ x = x - x_mean
+
+ U, S, VT = torch.linalg.svd( x )
+
+ #transpose 
+
+ #V = rearrange(VT, 'a b c -> a c b')
+ V = VT.permute( 0, 2, 1)
+ V = V[ :, :, 0 : 1 ]
+
+ projection = torch.bmm(x, V).squeeze( -1 )
+
+ #projection = rearrange( projection, 'b (h w) -> b h w', h = h, w = w)
+ projection = projection.reshape( b, h, w)
+
+ return projection.detach().numpy( )