Upload 4 files

monotonic_align/__init__.py

@@ -0,0 +1,15 @@
+from numpy import zeros, int32, float32
+from torch import from_numpy
+
+from .core import maximum_path_jit
+
+def maximum_path(neg_cent, mask):
+  device = neg_cent.device
+  dtype = neg_cent.dtype
+  neg_cent = neg_cent.data.cpu().numpy().astype(float32)
+  path = zeros(neg_cent.shape, dtype=int32)
+
+  t_t_max = mask.sum(1)[:, 0].data.cpu().numpy().astype(int32)
+  t_s_max = mask.sum(2)[:, 0].data.cpu().numpy().astype(int32)
+  maximum_path_jit(path, neg_cent, t_t_max, t_s_max)
+  return from_numpy(path).to(device=device, dtype=dtype)

monotonic_align/core.py

@@ -0,0 +1,35 @@
+import numba
+
+
+@numba.jit(numba.void(numba.int32[:,:,::1], numba.float32[:,:,::1], numba.int32[::1], numba.int32[::1]), nopython=True, nogil=True)
+def maximum_path_jit(paths, values, t_ys, t_xs):
+  b = paths.shape[0]
+  max_neg_val=-1e9
+  for i in range(int(b)):
+    path = paths[i]
+    value = values[i]
+    t_y = t_ys[i]
+    t_x = t_xs[i]
+
+    v_prev = v_cur = 0.0
+    index = t_x - 1
+
+    for y in range(t_y):
+      for x in range(max(0, t_x + y - t_y), min(t_x, y + 1)):
+        if x == y:
+          v_cur = max_neg_val
+        else:
+          v_cur = value[y-1, x]
+        if x == 0:
+          if y == 0:
+            v_prev = 0.
+          else:
+            v_prev = max_neg_val
+        else:
+          v_prev = value[y-1, x-1]
+        value[y, x] += max(v_prev, v_cur)
+
+    for y in range(t_y - 1, -1, -1):
+      path[y, index] = 1
+      if index != 0 and (index == y or value[y-1, index] < value[y-1, index-1]):
+        index = index - 1