`check_requirements()` is unreliable for large packages like torch and tensorflow that may have multiple installation routes (i.e. conda, pip, tensorflow-cpu, etc.)modifyDataloader
@@ -94,7 +94,6 @@ def run(weights=ROOT / 'yolov5s.pt', # model.pt path(s) | |||
import onnxruntime | |||
session = onnxruntime.InferenceSession(w, None) | |||
else: # TensorFlow models | |||
check_requirements(('tensorflow>=2.4.1',)) | |||
import tensorflow as tf | |||
if pb: # https://www.tensorflow.org/guide/migrate#a_graphpb_or_graphpbtxt | |||
def wrap_frozen_graph(gd, inputs, outputs): |