Oh, if thatās the case, I donāt know which one it is. Anyway, itās the default one.
I found out while searching earlier that, in the Colab environment, the pip contents are not overwritten unless we explicitly uninstall the libraryā¦
I used this as a base and cut and pasted the missing parts as I saw fit. Iām going to check out Colab now.
https://studiolab.sagemaker.aws/import/github/huggingface/notebooks/blob/main/transformers_doc/en/pytorch/semantic_segmentation.ipynb
I deleted the save section because it wasnāt needed for the test.
!pip install transformers datasets huggingface_hub==0.25.2
from datasets import load_dataset
ds = load_dataset("scene_parse_150", split="train[:50]", trust_remote_code=True)
ds = ds.train_test_split(test_size=0.2)
train_ds = ds["train"]
test_ds = ds["test"]
train_ds[0]
import json
from huggingface_hub import cached_download, hf_hub_url
repo_id = "huggingface/label-files"
filename = "ade20k-id2label.json"
id2label = json.load(open(cached_download(hf_hub_url(repo_id, filename, repo_type="dataset")), "r"))
id2label = {int(k): v for k, v in id2label.items()}
label2id = {v: k for k, v in id2label.items()}
num_labels = len(id2label)
from transformers import AutoImageProcessor
checkpoint = "nvidia/mit-b0"
image_processor = AutoImageProcessor.from_pretrained(checkpoint, reduce_labels=True)
from torchvision.transforms import ColorJitter
jitter = ColorJitter(brightness=0.25, contrast=0.25, saturation=0.25, hue=0.1)
def train_transforms(example_batch):
images = [jitter(x) for x in example_batch["image"]]
labels = [x for x in example_batch["annotation"]]
inputs = image_processor(images, labels)
return inputs
def val_transforms(example_batch):
images = [x for x in example_batch["image"]]
labels = [x for x in example_batch["annotation"]]
inputs = image_processor(images, labels)
return inputs
train_ds.set_transform(train_transforms)
test_ds.set_transform(val_transforms)
import evaluate
metric = evaluate.load("mean_iou")
import numpy as np
import torch
from torch import nn
def compute_metrics(eval_pred):
with torch.no_grad():
logits, labels = eval_pred
logits_tensor = torch.from_numpy(logits)
logits_tensor = nn.functional.interpolate(
logits_tensor,
size=labels.shape[-2:],
mode="bilinear",
align_corners=False,
).argmax(dim=1)
pred_labels = logits_tensor.detach().cpu().numpy()
metrics = metric.compute(
predictions=pred_labels,
references=labels,
num_labels=num_labels,
ignore_index=255,
reduce_labels=False,
)
for key, value in metrics.items():
if type(value) is np.ndarray:
metrics[key] = value.tolist()
return metrics
from transformers import AutoModelForSemanticSegmentation, TrainingArguments, Trainer
model = AutoModelForSemanticSegmentation.from_pretrained(checkpoint, id2label=id2label, label2id=label2id)
training_args = TrainingArguments(
output_dir="segformer-b0-scene-parse-150",
learning_rate=6e-5,
num_train_epochs=50,
per_device_train_batch_size=2,
per_device_eval_batch_size=2,
save_total_limit=3,
evaluation_strategy="steps",
save_strategy="steps",
save_steps=20,
eval_steps=20,
logging_steps=1,
eval_accumulation_steps=5,
remove_unused_columns=False,
push_to_hub=False,
)
trainer = Trainer(
model=model,
args=training_args,
train_dataset=train_ds,
eval_dataset=test_ds,
compute_metrics=compute_metrics,
)
trainer.train()
image = ds[0]["image"]
image
from transformers import pipeline
segmenter = pipeline("image-segmentation", model="my_awesome_seg_model")
segmenter(image)
Transformers. Use 