Flower Species Recognition -- 2
Budget: ₹1,500 – ₹12,500 INR
I’m building an end-to-end flower–image recognition system that runs as a web application. Users will be able to upload photos from their local drives, and the backend will instantly identify the flower species with high accuracy.
Here is the scope I need covered:
• Data pipeline – curate or expand an existing public flower dataset, handle cleaning, augmentation, and train/validation/test splits.
• Model development – implement a Convolutional Neural Network in Python (TensorFlow / Keras or PyTorch) tuned specifically for multi-class flower classification. I’m open to transfer-learning from ImageNet-based architectures if that speeds convergence.
• Evaluation – provide precision, recall, F1, and confusion matrix on the held-out test set, plus brief analysis of misclassifications.
• Web interface – a lightweight, responsive front end (Flask, FastAPI, or Django) where users choose an image file, trigger inference, and see the predicted species and confidence score. No camera or URL uploads are required—local file selection only.
• Deployment – Docker file or step-by-step guide so I can spin the service up on my own server (Ubuntu). Heroku or AWS instructions are a plus, but local hosting is the priority.
• Deliverables – fully commented source code, README with setup and usage steps, model weights, and a short technical document that explains the preprocessing workflow, architecture choices, and hyperparameters.
Acceptance criteria: the model must achieve at least 90 % top-1 accuracy on the test data, the web UI should return a result in under two seconds for a 512×512 input, and every instruction needed to reproduce the environment should be included.
Here is the scope I need covered:
• Data pipeline – curate or expand an existing public flower dataset, handle cleaning, augmentation, and train/validation/test splits.
• Model development – implement a Convolutional Neural Network in Python (TensorFlow / Keras or PyTorch) tuned specifically for multi-class flower classification. I’m open to transfer-learning from ImageNet-based architectures if that speeds convergence.
• Evaluation – provide precision, recall, F1, and confusion matrix on the held-out test set, plus brief analysis of misclassifications.
• Web interface – a lightweight, responsive front end (Flask, FastAPI, or Django) where users choose an image file, trigger inference, and see the predicted species and confidence score. No camera or URL uploads are required—local file selection only.
• Deployment – Docker file or step-by-step guide so I can spin the service up on my own server (Ubuntu). Heroku or AWS instructions are a plus, but local hosting is the priority.
• Deliverables – fully commented source code, README with setup and usage steps, model weights, and a short technical document that explains the preprocessing workflow, architecture choices, and hyperparameters.
Acceptance criteria: the model must achieve at least 90 % top-1 accuracy on the test data, the web UI should return a result in under two seconds for a 512×512 input, and every instruction needed to reproduce the environment should be included.