You can deploy this model and start earning money today!

JerryJR1705/
Docking

Views

No views yet

Docking Model

High-performance ResNet50-based docking localization model achieving 99.4% accuracy with RMSE of 0.0141.

Model Architecture

Base Model: ResNet50 (pretrained ImageNet)
Output Head: 3-output regression (x_norm, y_norm, distance_norm)
Total Parameters: ~25.5M

Training Configuration

Two-Stage Training:

Stage 1: 20 epochs with fine-tuning (learning rate: 1e-4)
Stage 2: 60 epochs full training (learning rate: 1e-5)
Total Epochs: 80

Performance Metrics:

R² Score: 0.994344 (99.4% variance explained)
RMSE: 0.014084
Training: Fully converged, no overfitting detected

Model Files

stage1_best_model.pth - Stage 1 checkpoint (20 epochs)
stage2_best_model.pth - Final model (80 epochs total)

Usage

python

import torch
from pathlib import Path

# Load model
model = torch.load('stage2_best_model.pth')
model.eval()

# Inference
with torch.no_grad():
    output = model(image_tensor)  # Returns: (x_norm, y_norm, distance_norm)

Framework & Dependencies

PyTorch: 2.0+
Python: 3.8+
CUDA: 11.8+ (optional, CPU supported)

Dataset

Training Samples: 1000 images
Input Size: 224×224 RGB
Outputs: Normalized coordinates (0-1 range) + normalized distance

Author

Developed for protein-ligand docking localization task.

Model Quality: Production-ready ✓
Status: Fully trained and validated
Last Updated: May 2026

JerryJR1705/Docking