resnet分类模型无法指定分类数量

## 使用 classification resnet进行分类苹果/橘子训练时，指定分类数量为2时，无法训练
Models/official/vision/classification/resnet/model.py
resnet50只支持1000分类数据集训练，我只创建2分类数据时，修改num_classes=2时，无法正常完成训练

## 任务描述
数据集格式是这样的：
/path/to/imagenet
train
apple
xxx.jpg
...
orangle
xxx.jpg
...
...
val
apple
xxx.jpg
...
orangle
xxx.jpg
...

## 目标
由于我时初学，在经过自己各种修改后，仍然无法解决，需要帮忙。谢谢

Start in Models/official/vision/classification/resnet/model.py and reproduce training with the shown train/val directory layout and num_classes=2 setting. The issue does not include an error message or failing test, so first capture the training failure; done means the ResNet model trains successfully on the two-class apple/orange dataset.

Written by the indexing model from the issue text.