Fix

app.py

@@ -1,13 +1,13 @@
 import os
 from typing import List, Tuple
 import multiprocessing
+import cv2
 
 import numpy as np
 import pandas as pd
 import streamlit as st
 import torch
 from torch import Tensor
-from decord import VideoReader, cpu
 from transformers import AutoFeatureExtractor, TimesformerForVideoClassification
 
 np.random.seed(0)
@@ -49,7 +49,27 @@ def load_model(model_name: str):
     return feature_extractor, model
 
 
-def inference(file_path: str):
+def read_video(file_path: str) -> np.ndarray:
+    cap = cv2.VideoCapture(file_path)
+    length = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))  # 1000 frames
+    print("Number of frames", length)
+
+    indices = sample_frame_indices(clip_len=8, frame_sample_rate=4, seg_len=length)
+
+    frames: List[np.array] = []
+    for i in indices:
+        cap.set(1, i)
+        ret, frame = cap.read()
+        if not ret:
+            continue
+        frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+        frames.append(frame)
+    return np.array(frames)
+
+
+def read_video_decord(file_path: str) -> np.ndarray:
+    from decord import VideoReader, cpu
+
     videoreader = VideoReader(VIDEO_TMP_PATH, num_threads=1, ctx=cpu(0))
 
     # sample 8 frames
@@ -58,6 +78,12 @@ def inference(file_path: str):
         clip_len=8, frame_sample_rate=4, seg_len=len(videoreader)
     )
     video = videoreader.get_batch(indices).asnumpy()
+    # print(video.shape) # (8, 720, 1280, 3)
+    return video
+
+
+def inference(file_path: str):
+    video = read_video(file_path)
 
     inputs = feature_extractor(list(video), return_tensors="pt")
 

requirements.txt

@@ -1,5 +1,6 @@
 streamlit
 transformers
 torch
-decord
+# decord
 black
+opencv-python