Finetune Instance Segmentation Ade20k Mini Mask2former No Trainer

🚀 Instance Segmentation Example

This project provides an example of instance segmentation, including training and inference steps.

🚀 Quick Start

This README guides you through the process of instance segmentation, including training the model and performing inference.

✨ Features

• Multi - environment Support: The training script uses 🤗 Accelerate to support various environments such as CPU, multi - CPU, GPU, multi - GPU, and TPU, with mixed - precision support.
• Easy Inference: You can easily load the trained model and perform inference on new images.

📦 Installation

Before starting, you need to configure the 🤗 Accelerate environment:

accelerate config

Answer the questions according to your training environment. Then, run the following command to ensure everything is ready for training:

accelerate test

💻 Usage Examples

Basic Usage

Training the Model

To train the instance segmentation model, use the following command:

accelerate launch run_instance_segmentation_no_trainer.py \
    --model_name_or_path facebook/mask2former-swin-tiny-coco-instance \
    --output_dir finetune-instance-segmentation-ade20k-mini-mask2former-no-trainer \
    --dataset_name qubvel-hf/ade20k-mini \
    --do_reduce_labels \
    --image_height 256 \
    --image_width 256 \
    --num_train_epochs 40 \
    --learning_rate 1e-5 \
    --lr_scheduler_type constant \
    --per_device_train_batch_size 8 \
    --gradient_accumulation_steps 2 \
    --dataloader_num_workers 8 \
    --push_to_hub

Performing Inference

You can load the trained model and perform inference on an image as follows:

import torch
import requests
import matplotlib.pyplot as plt

from PIL import Image
from transformers import Mask2FormerForUniversalSegmentation, Mask2FormerImageProcessor

# Load image
image = Image.open(requests.get("http://farm4.staticflickr.com/3017/3071497290_31f0393363_z.jpg", stream=True).raw)

# Load model and image processor
device = "cuda"
checkpoint = "qubvel-hf/finetune-instance-segmentation-ade20k-mini-mask2former-no-trainer"

model = Mask2FormerForUniversalSegmentation.from_pretrained(checkpoint, device_map=device)
image_processor = Mask2FormerImageProcessor.from_pretrained(checkpoint)

# Run inference on image
inputs = image_processor(images=[image], return_tensors="pt").to(device)
with torch.no_grad():
    outputs = model(**inputs)

# Post-process outputs
outputs = image_processor.post_process_instance_segmentation(outputs, target_sizes=[image.size[::-1]])

print("Mask shape: ", outputs[0]["segmentation"].shape)
print("Mask values: ", outputs[0]["segmentation"].unique())
for segment in outputs[0]["segments_info"]:
    print("Segment: ", segment)

Advanced Usage

Visualizing the Results

Use the following code to visualize the instance segmentation results:

import numpy as np
import matplotlib.pyplot as plt

segmentation = outputs[0]["segmentation"].numpy()

plt.figure(figsize=(10, 10))
plt.subplot(1, 2, 1)
plt.imshow(np.array(image))
plt.axis("off")
plt.subplot(1, 2, 2)
plt.imshow(segmentation)
plt.axis("off")
plt.show()

📄 License

Copyright 2024 The HuggingFace Team. All rights reserved.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.