# 'ViTImageProcessor' object has no attribute 'pad' **URL:** \ **Category:** Beginners\ **Created:** [February 25, 2023, 4:43pm UTC](https://discuss.huggingface.co/t/vitimageprocessor-object-has-no-attribute-pad/32511 "2023-02-25T16:43:00Z")\ **Posts on this page:** 5\ **Page:** 1
**Author:** ![yoyo1a1](https://avatars.discourse-cdn.com/v4/letter/y/f475e1/32.png) [@yoyo1a1](https://discuss.huggingface.co/u/yoyo1a1)\ **Post date:** [February 25, 2023, 4:43pm UTC](https://discuss.huggingface.co/t/vitimageprocessor-object-has-no-attribute-pad/32511/1 "2023-02-25T16:43:01Z")
Hi I’m a student, new to this topic in general and that’s my first time using Hugging Face. I want to fine-tune ViT with my own dataset, In order to do so, I’m following [this tutorial](https://huggingface.co/blog/fine-tune-vit), but since I don’t import my data set from datasets, some things are different. This is my custom data set class: import os import torch import numpy as np from transformers import ViTImageProcessor from torchvision import transforms from PIL import Image model\_name\_or\_path = ‘google/vit-base-patch16-224-in21k’ feature\_extractor = ViTImageProcessor.from\_pretrained(model\_name\_or\_path) ```auto class ViTCustomDataSet(): def __init__ (self, image_dir, device): # super(ImageDataset, self). __init__ () self.image_dir = image_dir self.transform = transforms.Compose([ transforms.ToPILImage(), self.extract_features ]) self.image_dir = image_dir self.device = device def extract_features(self, image): return feature_extractor(image, return_tensors='pt') return feature_extractor(image, return_tensors='pt')['pixel_values'] def __len__ (self): return len(os.listdir(self.image_dir)) def __getitem__ (self, idx): image_name = os.listdir(self.image_dir)[idx] image_path = os.path.join(self.image_dir, image_name) image = np.array(Image.open(image_path).convert('RGB')) image = self.transform(image) label = self.get_label_from_filename(image_name) return image.to(self.device), label def get_label_from_filename(self, image_name): return 1 if image_name.split('_')[-2] == 'M' else 0 def get_train_and_test_data(self): train_size = int(0.8 * self. __len__ ()) test_size = self. __len__ () - train_size train_dataset, test_dataset = torch.utils.data.random_split( self, [train_size, test_size]) return torch.utils.data.DataLoader( train_dataset, batch_size=32, shuffle=True).dataset, torch.utils.data.DataLoader( test_dataset, batch_size=32, shuffle=True).dataset ``` And that’s my main: ```auto training_args = TrainingArguments( output_dir="./vit", # per_device_train_batch_size=16, evaluation_strategy="steps", num_train_epochs=4, # fp16=True, save_steps=100, eval_steps=100, logging_steps=10, learning_rate=2e-4, save_total_limit=2, remove_unused_columns=False, push_to_hub=False, report_to='tensorboard', load_best_model_at_end=True, ) device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu') vitData = ViTCustomDataSet( 'D:/data/NIST302/images/auxiliary/flat/M/500/plain/png/equal', device=device) train_dataset, test_dataset = vitData.get_train_and_test_data() model = ViTForImageClassification.from_pretrained( model_name_or_path, num_labels=num_classes) feature_extractor = ViTImageProcessor.from_pretrained( model_name_or_path, padding=False) trainer = Trainer( model=model, args=training_args, compute_metrics=compute_metrics, train_dataset=train_dataset, eval_dataset=test_dataset, data_collator=None, tokenizer=feature_extractor, ) train_results = trainer.train() trainer.save_model() trainer.log_metrics("train", train_results.metrics) trainer.save_metrics("train", train_results.metrics) trainer.save_state() metrics = trainer.evaluate(test_dataset) trainer.log_metrics("eval", metrics) trainer.save_metrics("eval", metrics) ``` (I’m sorry for the mess, I’m transforming an old code of mine and I want to make sure things work before I refactor) Anyway, I’m getting the following error: > ViTImageProcessor does not have a pad attribute Which I don’t see how to solve. Also - since I’m new, any other tip will be welcomed! Thank you ---
**Author:** ![iwaitu](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/iwaitu/32/29935_2.png) [@iwaitu](https://discuss.huggingface.co/u/iwaitu)\ **Post date:** [September 27, 2023, 4:55pm UTC](https://discuss.huggingface.co/t/vitimageprocessor-object-has-no-attribute-pad/32511/2 "2023-09-27T16:55:52Z")
I have the same problem. ---
**Author:** ![Sandy1857](https://avatars.discourse-cdn.com/v4/letter/s/ecae2f/32.png) [@Sandy1857](https://discuss.huggingface.co/u/Sandy1857)\ **Post date:** [October 6, 2023, 12:53pm UTC](https://discuss.huggingface.co/t/vitimageprocessor-object-has-no-attribute-pad/32511/3 "2023-10-06T12:53:30Z")
Just remove the padding parameter from the below: ```auto feature_extractor = ViTImageProcessor.from_pretrained( model_name_or_path, padding=False) ``` It should be just: ```auto feature_extractor = ViTImageProcessor.from_pretrained( model_name_or_path) ``` ---
**Author:** ![minhocas](https://avatars.discourse-cdn.com/v4/letter/m/cdc98d/32.png) [@minhocas](https://discuss.huggingface.co/u/minhocas)\ **Post date:** [December 1, 2023, 5:16pm UTC](https://discuss.huggingface.co/t/vitimageprocessor-object-has-no-attribute-pad/32511/4 "2023-12-01T17:16:06Z")
Did you resolve this issue? I’m having the same problem. ---
**Author:** ![nielsr](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/nielsr/32/39617_2.png) [@nielsr](https://discuss.huggingface.co/u/nielsr)\ **Post date:** [March 27, 2024, 8:08am UTC](https://discuss.huggingface.co/t/vitimageprocessor-object-has-no-attribute-pad/32511/5 "2024-03-27T08:08:21Z")
Hi, This can be bypassed by passing the default data collator to the Trainer: ```auto from transformers import default_data_collator trainer = Trainer(... tokenizer=image_processor, data_collator=default_data_collator ) ``` However, we’re going to fix this issue by adding an `image_processor` argument to the Trainer so that users no longer need to do this. You can follow the progress here: [[Trainer] Allow passing image processor by NielsRogge · Pull Request #29896 · huggingface/transformers · GitHub](https://github.com/huggingface/transformers/pull/29896)