MindSpore的VIT报错[OneHot] failed. OneHot: index values should not bigger than num classes: 100, but got: 100.及解决

[huan666]

1 系统环境

硬件环境(Ascend/GPU/CPU): Ascend/GPU/CPU
MindSpore版本: mindspore=2.0.0
执行模式（PyNative/ Graph）:不限
Python版本: Python=3.9
操作系统平台: 不限

2 报错信息

2.1 问题描述

使用Imagenet数据集，设置了100个类型的图片分类。

{'autoaugment': 1,   
 'backbone': 'vit_base_patch32',   
 'batch_size': 8,   
 'beta1': 0.9,   
 'beta2': 0.999,   
 'class_num': 100,   
 'config_path': 'vit_patch32_imagenet2012_config_standalone.yml',   
 'crop_min': 0.05,   
 'dataset_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/train',   
 'enable_modelarts': 0,   
 'eval_batch_size': 4,   
 'eval_image_size': 224,   
 'eval_interval': 1,   
 'eval_num_workers': 2,   
 'eval_offset': -1,   
 'eval_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/val',   
 'gc_flag': 0,   
 'interpolation': 'BILINEAR',   
 'keep_checkpoint_max': 3,   
 'label_smooth_factor': 0.1,   
 'loss_name': 'ce_smooth_mixup',   
 'loss_scale': 1024,   
 'lr_decay_mode': 'cosine',   
 'lr_init': 0.001,   
 'lr_max': 0.00044375,   
 'lr_min': 1e-08,   
 'max_epoch': 300,   
 'mixup': 0.2,   
 'no_weight_decay_filter': 'beta,bias',   
 'open_profiler': 0,   
 'opt': 'adamw',   
 'pretrained': '',   
 'save_checkpoint': 1,   
 'save_checkpoint_epochs': 8,   
 'save_checkpoint_path': './outputs',   
 'train_image_size': 224,   
 'train_num_workers': 1,   
 'use_label_smooth': 1,   
 'vit_config_path': 'src.vit.VitConfig',   
 'warmup_epochs': 0,   
 'weight_decay': 0.05}   
Please check the above information for the configurations   
standalone training   
########################################   
{'auto_tune': 0,   
 'autoaugment': 1,   
 'aux_factor': 0.4,   
 'backbone': 'vit_base_patch32',   
 'batch_size': 8,   
 'beta1': 0.9,   
 'beta2': 0.999,   
 'class_num': 100,   
 'config_path': 'vit_patch32_imagenet2012_config_standalone.yml',   
 'crop_min': 0.05,   
 'dataset_name': 'imagenet',   
 'dataset_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/train',   
 'device_id': 0,   
 'device_num': 1,   
 'enable_modelarts': 0,   
 'eval_batch_size': 4,   
 'eval_engine': 'imagenet',   
 'eval_image_size': 224,   
 'eval_interval': 1,   
 'eval_num_workers': 2,   
 'eval_offset': 0,   
 'eval_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/val',   
 'gc_flag': 0,   
 'interpolation': 'BILINEAR',   
 'keep_checkpoint_max': 3,   
 'label_smooth_factor': 0.1,   
 'local_rank': 0,   
 'logger': <LOGGER mindspore-benchmark (NOTSET)>,   
 'loss_name': 'ce_smooth_mixup',   
 'loss_scale': 1024,   
 'lr_decay_mode': 'cosine',   
 'lr_init': 0.001,   
 'lr_max': 0.00044375,   
 'lr_min': 1e-08,   
 'max_epoch': 300,   
 'mixup': 0.2,   
 'no_weight_decay_filter': 'beta,bias',   
 'open_profiler': 0,   
 'opt': 'adamw',   
 'poly_power': 2,   
 'pretrained': '',   
 'save_checkpoint': 1,   
 'save_checkpoint_epochs': 8,   
 'save_checkpoint_path': './outputs',   
 'seed': 1,   
 'split_point': 0.4,   
 'train_image_size': 224,   
 'train_num_workers': 1,   
 'use_label_smooth': 1,   
 'vit_config_path': 'src.vit.VitConfig',   
 'warmup_epochs': 0,   
 'weight_decay': 0.05}   
########################################   
get vit_config_path   
split_point=0.4   
============================================================   
{'auto_tune': 0,   
 'autoaugment': 1,   
 'aux_factor': 0.4,   
 'backbone': 'vit_base_patch32',   
 'batch_size': 8,   
 'beta1': 0.9,   
 'beta2': 0.999,   
 'class_num': 100,   
 'config_path': 'vit_patch32_imagenet2012_config_standalone.yml',   
 'crop_min': 0.05,   
 'dataset_name': 'imagenet',   
 'dataset_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/train',   
 'device_id': 0,   
 'device_num': 1,   
 'enable_modelarts': 0,   
 'eval_batch_size': 4,   
 'eval_engine': 'imagenet',   
 'eval_image_size': 224,   
 'eval_interval': 1,   
 'eval_num_workers': 2,   
 'eval_offset': 0,   
 'eval_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/val',   
 'gc_flag': 0,   
 'interpolation': 'BILINEAR',   
 'keep_checkpoint_max': 3,   
 'label_smooth_factor': 0.1,   
 'local_rank': 0,   
 'logger': <LOGGER mindspore-benchmark (NOTSET)>,   
 'loss_name': 'ce_smooth_mixup',   
 'loss_scale': 1024,   
 'lr_decay_mode': 'cosine',   
 'lr_init': 0.001,   
 'lr_max': 0.00044375,   
 'lr_min': 1e-08,   
 'max_epoch': 300,   
 'mixup': 0.2,   
 'no_weight_decay_filter': 'beta,bias',   
 'open_profiler': 0,   
 'opt': 'adamw',   
 'poly_power': 2,   
 'pretrained': '',   
 'save_checkpoint': 1,   
 'save_checkpoint_epochs': 8,   
 'save_checkpoint_path': './outputs',   
 'seed': 1,   
 'split_point': 0.4,   
 'train_image_size': 224,   
 'train_num_workers': 1,   
 'use_label_smooth': 1,   
 'vit_config_path': 'src.vit.VitConfig',   
 'warmup_epochs': 0,   
 'weight_decay': 0.05}   
============================================================ 

3 根因分析

imagenet数据集是class=100，但是0-100表示的是101个类，不是100个类，所以需要设置成0-99。

4 解决方案

将代码中的’class_num’: 100,修改为’class_num’: 99.

{'autoaugment': 1,   
 'backbone': 'vit_base_patch32',   
 'batch_size': 8,   
 'beta1': 0.9,   
 'beta2': 0.999,   
 'class_num': 99,   
 'config_path': 'vit_patch32_imagenet2012_config_standalone.yml',   
 'crop_min': 0.05,   
 'dataset_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/train',   
 'enable_modelarts': 0,   
 'eval_batch_size': 4,   
 'eval_image_size': 224,   
 'eval_interval': 1,   
 'eval_num_workers': 2,   
 'eval_offset': -1,   
 'eval_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/val',   
 'gc_flag': 0,   
 'interpolation': 'BILINEAR',   
 'keep_checkpoint_max': 3,   
 'label_smooth_factor': 0.1,   
 'loss_name': 'ce_smooth_mixup',   
 'loss_scale': 1024,   
 'lr_decay_mode': 'cosine',   
 'lr_init': 0.001,   
 'lr_max': 0.00044375,   
 'lr_min': 1e-08,   
 'max_epoch': 300,   
 'mixup': 0.2,   
 'no_weight_decay_filter': 'beta,bias',   
 'open_profiler': 0,   
 'opt': 'adamw',   
 'pretrained': '',   
 'save_checkpoint': 1,   
 'save_checkpoint_epochs': 8,   
 'save_checkpoint_path': './outputs',   
 'train_image_size': 224,   
 'train_num_workers': 1,   
 'use_label_smooth': 1,   
 'vit_config_path': 'src.vit.VitConfig',   
 'warmup_epochs': 0,   
 'weight_decay': 0.05}   
Please check the above information for the configurations   
standalone training   
########################################   
{'auto_tune': 0,   
 'autoaugment': 1,   
 'aux_factor': 0.4,   
 'backbone': 'vit_base_patch32',   
 'batch_size': 8,   
 'beta1': 0.9,   
 'beta2': 0.999,   
 'class_num': 99,   
 'config_path': 'vit_patch32_imagenet2012_config_standalone.yml',   
 'crop_min': 0.05,   
 'dataset_name': 'imagenet',   
 'dataset_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/train',   
 'device_id': 0,   
 'device_num': 1,   
 'enable_modelarts': 0,   
 'eval_batch_size': 4,   
 'eval_engine': 'imagenet',   
 'eval_image_size': 224,   
 'eval_interval': 1,   
 'eval_num_workers': 2,   
 'eval_offset': 0,   
 'eval_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/val',   
 'gc_flag': 0,   
 'interpolation': 'BILINEAR',   
 'keep_checkpoint_max': 3,   
 'label_smooth_factor': 0.1,   
 'local_rank': 0,   
 'logger': <LOGGER mindspore-benchmark (NOTSET)>,   
 'loss_name': 'ce_smooth_mixup',   
 'loss_scale': 1024,   
 'lr_decay_mode': 'cosine',   
 'lr_init': 0.001,   
 'lr_max': 0.00044375,   
 'lr_min': 1e-08,   
 'max_epoch': 300,   
 'mixup': 0.2,   
 'no_weight_decay_filter': 'beta,bias',   
 'open_profiler': 0,   
 'opt': 'adamw',   
 'poly_power': 2,   
 'pretrained': '',   
 'save_checkpoint': 1,   
 'save_checkpoint_epochs': 8,   
 'save_checkpoint_path': './outputs',   
 'seed': 1,   
 'split_point': 0.4,   
 'train_image_size': 224,   
 'train_num_workers': 1,   
 'use_label_smooth': 1,   
 'vit_config_path': 'src.vit.VitConfig',   
 'warmup_epochs': 0,   
 'weight_decay': 0.05}   
########################################   
get vit_config_path   
split_point=0.4   
============================================================   
{'auto_tune': 0,   
 'autoaugment': 1,   
 'aux_factor': 0.4,   
 'backbone': 'vit_base_patch32',   
 'batch_size': 8,   
 'beta1': 0.9,   
 'beta2': 0.999,   
 'class_num': 99,   
 'config_path': 'vit_patch32_imagenet2012_config_standalone.yml',   
 'crop_min': 0.05,   
 'dataset_name': 'imagenet',   
 'dataset_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/train',   
 'device_id': 0,   
 'device_num': 1,   
 'enable_modelarts': 0,   
 'eval_batch_size': 4,   
 'eval_engine': 'imagenet',   
 'eval_image_size': 224,   
 'eval_interval': 1,   
 'eval_num_workers': 2,   
 'eval_offset': 0,   
 'eval_path': '/data/zx_cv/vit/models-master/VIT/cache/datasets/imagenet_cifa/tiny-new/datasets/val',   
 'gc_flag': 0,   
 'interpolation': 'BILINEAR',   
 'keep_checkpoint_max': 3,   
 'label_smooth_factor': 0.1,   
 'local_rank': 0,   
 'logger': <LOGGER mindspore-benchmark (NOTSET)>,   
 'loss_name': 'ce_smooth_mixup',   
 'loss_scale': 1024,   
 'lr_decay_mode': 'cosine',   
 'lr_init': 0.001,   
 'lr_max': 0.00044375,   
 'lr_min': 1e-08,   
 'max_epoch': 300,   
 'mixup': 0.2,   
 'no_weight_decay_filter': 'beta,bias',   
 'open_profiler': 0,   
 'opt': 'adamw',   
 'poly_power': 2,   
 'pretrained': '',   
 'save_checkpoint': 1,   
 'save_checkpoint_epochs': 8,   
 'save_checkpoint_path': './outputs',   
 'seed': 1,   
 'split_point': 0.4,   
 'train_image_size': 224,   
 'train_num_workers': 1,   
 'use_label_smooth': 1,   
 'vit_config_path': 'src.vit.VitConfig',   
 'warmup_epochs': 0,   
 'weight_decay': 0.05}   
============================================================