---
language:
- en
license: mit
library_name: pytorch
tags:
- computer-vision
- object-detection
- fcos
- minecraft
- gaming
datasets:
- twoturtles/minecraft-mobs
metrics:
- mean_average_precision
---

# FCOS Minecraft Mob Detector

Detects 9 Minecraft entities (chicken, cow, creeper, enderman, pig, sheep, skeleton, spider, zombie) using FCOS (Fully Convolutional One-Stage) object detection.

## Code

For training details and usage see [minecraft-fcos](https://github.com/twoturtles/minecraft-fcos)

The code allows you to run the model over Minecraft in realtime.



## Model Details

- **Architecture**: FCOS with ResNet50-FPN backbone
- **Input**: 640x640 RGB images
- **Output**: Bounding boxes and class labels for 9 entity types
- **Training**: Transfer learning on FCOS-pretrained weights
- **Framework**: PyTorch + torchvision

## Performance

| Metric | Value |
|--------|-------|
| mAP | 0.595 |
| mAP_50 | 0.84 |

## Sample Detection

![detection](https://cdn-uploads.huggingface.co/production/uploads/690be6f579dee59c823691db/lfLZ0_C1VdpzFMdfVU7o_.png)