Migrate metric to Evaluate in Pytorch examples (#18369)

* Migrate metric to Evaluate in pytorch examples

* Remove unused imports

examples/pytorch/semantic-segmentation/run_semantic_segmentation.py

@@ -21,7 +21,6 @@ import sys
 from dataclasses import dataclass, field
 from typing import Optional
 
-import datasets
 import numpy as np
 import torch
 from datasets import load_dataset
@@ -30,6 +29,7 @@ from torch import nn
 from torchvision import transforms
 from torchvision.transforms import functional
 
+import evaluate
 import transformers
 from huggingface_hub import hf_hub_download
 from transformers import (
@@ -337,7 +337,7 @@ def main():
     label2id = {v: str(k) for k, v in id2label.items()}
 
     # Load the mean IoU metric from the datasets package
-    metric = datasets.load_metric("mean_iou")
+    metric = evaluate.load("mean_iou")
 
     # Define our compute_metrics function. It takes an `EvalPrediction` object (a namedtuple with a
     # predictions and label_ids field) and has to return a dictionary string to float.