Switch from np.frombuffer to torch.frombuffer (#4578)

Co-authored-by: Prabhat Roy <prabhatroy@fb.com>

test/test_image.py

@@ -476,9 +476,7 @@ def test_encode_jpeg(img_path):
     buf = io.BytesIO()
     pil_img.save(buf, format="JPEG", quality=75)
 
-    # pytorch can't read from raw bytes so we go through numpy
-    pil_bytes = np.frombuffer(buf.getvalue(), dtype=np.uint8)
-    encoded_jpeg_pil = torch.as_tensor(pil_bytes)
+    encoded_jpeg_pil = torch.frombuffer(buf.getvalue(), dtype=torch.uint8)
 
     for src_img in [img, img.contiguous()]:
         encoded_jpeg_torch = encode_jpeg(src_img, quality=75)

test/test_video_reader.py

@@ -258,7 +258,7 @@ def _get_video_tensor(video_dir, video_file):
     assert os.path.exists(full_path), "File not found: %s" % full_path
 
     with open(full_path, "rb") as fp:
-        video_tensor = torch.from_numpy(np.frombuffer(fp.read(), dtype=np.uint8))
+        video_tensor = torch.frombuffer(fp.read(), dtype=torch.uint8)
 
     return full_path, video_tensor
 

torchvision/io/_video_opt.py

@@ -3,7 +3,6 @@ import warnings
 from fractions import Fraction
 from typing import List, Tuple
 
-import numpy as np
 import torch
 
 from .._internally_replaced_utils import _get_extension_path
@@ -338,7 +337,7 @@ def _read_video_from_memory(
     _validate_pts(audio_pts_range)
 
     if not isinstance(video_data, torch.Tensor):
-        video_data = torch.from_numpy(np.frombuffer(video_data, dtype=np.uint8))
+        video_data = torch.frombuffer(video_data, dtype=torch.uint8)
 
     result = torch.ops.video_reader.read_video_from_memory(
         video_data,
@@ -378,7 +377,7 @@ def _read_video_timestamps_from_memory(video_data):
     is much faster than read_video(...)
     """
     if not isinstance(video_data, torch.Tensor):
-        video_data = torch.from_numpy(np.frombuffer(video_data, dtype=np.uint8))
+        video_data = torch.frombuffer(video_data, dtype=torch.uint8)
     result = torch.ops.video_reader.read_video_from_memory(
         video_data,
         0,  # seek_frame_margin
@@ -415,7 +414,7 @@ def _probe_video_from_memory(video_data):
     This function is torchscriptable
     """
     if not isinstance(video_data, torch.Tensor):
-        video_data = torch.from_numpy(np.frombuffer(video_data, dtype=np.uint8))
+        video_data = torch.frombuffer(video_data, dtype=torch.uint8)
     result = torch.ops.video_reader.probe_video_from_memory(video_data)
     vtimebase, vfps, vduration, atimebase, asample_rate, aduration = result
     info = _fill_info(vtimebase, vfps, vduration, atimebase, asample_rate, aduration)