Don't cache input shapes because they can vary

Input shape transforms were previously only calculated once (i.e. for
one frame / one image) and then cached for all subsequent calls.
This is fine in 99% of use cases but if you want to process differently
shaped inputs in the same deface process (given multiple input paths),
this leads to crashes or even silently incorrect outputs.

Transforms now have to be recalculated for each frame but this is
very cheap, so it's okay to do this for the sake of better stability.

Fixes #41

deface/centerface.py

@@ -94,14 +94,13 @@ def dynamicize_shapes(static_model):
 
     def __call__(self, img, threshold=0.5):
         img = ensure_rgb(img)
-        self.orig_shape = img.shape[:2]
-        if self.in_shape is None:
-            self.in_shape = self.orig_shape[::-1]
-        if not hasattr(self, 'h_new'):  # First call, need to compute sizes
-            self.w_new, self.h_new, self.scale_w, self.scale_h = self.transform(self.in_shape)
+        orig_shape = img.shape[:2]
+        in_shape = orig_shape[::-1] if self.in_shape is None else self.in_shape
+        # Compute sizes
+        w_new, h_new, scale_w, scale_h = self.shape_transform(in_shape, orig_shape)
 
         blob = cv2.dnn.blobFromImage(
-            img, scalefactor=1.0, size=(self.w_new, self.h_new),
+            img, scalefactor=1.0, size=(w_new, h_new),
             mean=(0, 0, 0), swapRB=False, crop=False
         )
         if self.backend == 'opencv':
@@ -111,18 +110,19 @@ def __call__(self, img, threshold=0.5):
             heatmap, scale, offset, lms = self.sess.run(self.onnx_output_names, {self.onnx_input_name: blob})
         else:
             raise RuntimeError(f'Unknown backend {self.backend}')
-        dets, lms = self.decode(heatmap, scale, offset, lms, (self.h_new, self.w_new), threshold=threshold)
+        dets, lms = self.decode(heatmap, scale, offset, lms, (h_new, w_new), threshold=threshold)
         if len(dets) > 0:
-            dets[:, 0:4:2], dets[:, 1:4:2] = dets[:, 0:4:2] / self.scale_w, dets[:, 1:4:2] / self.scale_h
-            lms[:, 0:10:2], lms[:, 1:10:2] = lms[:, 0:10:2] / self.scale_w, lms[:, 1:10:2] / self.scale_h
+            dets[:, 0:4:2], dets[:, 1:4:2] = dets[:, 0:4:2] / scale_w, dets[:, 1:4:2] / scale_h
+            lms[:, 0:10:2], lms[:, 1:10:2] = lms[:, 0:10:2] / scale_w, lms[:, 1:10:2] / scale_h
         else:
             dets = np.empty(shape=[0, 5], dtype=np.float32)
             lms = np.empty(shape=[0, 10], dtype=np.float32)
 
         return dets, lms
 
-    def transform(self, in_shape):
-        h_orig, w_orig = self.orig_shape
+    @staticmethod
+    def shape_transform(in_shape, orig_shape):
+        h_orig, w_orig = orig_shape
         w_new, h_new = in_shape
         # Make spatial dims divisible by 32
         w_new, h_new = int(np.ceil(w_new / 32) * 32), int(np.ceil(h_new / 32) * 32)