I will train custom image segmentation and classification models
Machine Learning Engineer and Computer Vision Specialist
About this Gig
Need to outline regions in your images or predict a category for each image?
I will train a custom image segmentation or image classification model on your labeled dataset.
For semantic segmentation, I can train a PyTorch model using U-Net or DeepLabV3+ to generate pixel-level masks. If you need a separate mask for each object, I can scope instance segmentation based on your annotations.
For image classification, I use transfer learning with models such as ResNet or EfficientNet to predict categories like product types or defect classes.
You'll receive trained weights, Python source code, a runnable inference script, sample predictions, and setup instructions. I report task-appropriate metrics on held-out images, such as IoU or Dice for masks and per-class F1 for classification.
Send a few sample images, your labels, and the output you need. Ill confirm which package fits.
APIs:
Microsoft Computer Vision AI
•
Google Cloud Vision API
Programming language:
Python
•
SQL
•
Colab
•
Java
•
MLflow
Tools:
Jupyter Notebook
•
OpenCV
•
OpenNN
•
TensorFlow
•
CVAT
•
Colab
•
PyTorch
Frameworks:
Scikit-learn
•
SimpleCV
•
Keras
•
PyTorch
•
Panda
My Portfolio
Other Data Science & ML Services I Offer
FAQ
Can I order just segmentation or just classification?
Yes. Basic and Standard each cover one task. Choose segmentation for pixel-level masks or classification for image-level category predictions.
What labels do I need to provide?
For segmentation, share images with matching masks or polygon annotations. For classification, share labeled folders or a file mapping each image to its class. If you need both outputs, share both types of labels.
Can you work with unlabeled images?
Send a small sample and the approximate image count. I’ll check whether a pretrained demo is suitable or confirm what annotation is needed for custom training.
What results can I expect?
Performance depends on the dataset and label quality. I report results on held-out images and show examples of predictions, including common errors where applicable.
Can the same workflow show where a defect is and identify its type?
Yes. For example, the pipeline can produce a mask showing the defect area and an image-level defect label. Share your samples and annotations so I can confirm the scope.
Will I get the code and model weights?
Yes. Each package includes the trained weights, Python source code, an inference script, and instructions to run a sample prediction.

