MihaiPopa-1/minecraft-skins-1.1m-deduped-64x64-1.1

---
task_categories:
- text-to-image
- image-classification
- unconditional-image-generation
tags:
- minecraft
- minecraft-skins
- de-duped
- deduped
- zip-dataset
- zip
- zip-archive
size_categories:
- 1M<n<10M
license: apache-2.0
pretty_name: Minecraft Skins 1.1M Deduped (64x64 Edition) 1.1
---
# Minecraft Skins 1.1M Deduped (64x64 Edition) 1.1!
[Minecraft Skins 1.1M Deduped](https://huggingface.co/datasets/MihaiPopa-1/minecraft-skins-1.1m-deduped-64x64) but troll skins were filtered.

Format is a 2.6 GB ZIP archive containing 64x64 PNG skin files.

# Tools used
PIL Image (Python) and Google Colab (free CPU tier)

# How it was made
1. Loaded [Minecraft Skins 1.1M Deduped](https://huggingface.co/datasets/MihaiPopa-1/minecraft-skins-1.1m-deduped-64x64),
2. Troll (like single-color) skins were filtered (std > 1 were kept) and removed.
3. Result is 1106552 real 64x64 skins.
4. Output is given in a 2.6 GB ZIP archive.

This can be used to make your own skin generation model (but I'm going with VQ-VAE anyway!)

# Future improvements for version 2
1. Captioning (with Florence 2 Base)
2. ~~Filtering troll skins (skins that are formed of just a single color)~~ already done!

# Code
Code to reproduce it (all by Claude 4.6 Sonnet):

Same code as before, then:
```python
import os
import numpy as np
from PIL import Image
from tqdm import tqdm
from multiprocessing import Pool, cpu_count
import shutil

SKIN_DIR = "/content/filtered_skins"
OUTPUT_DIR = "/content/filtered_skins_plus"
STD_THRESHOLD = 1

os.makedirs(OUTPUT_DIR, exist_ok=True)
skin_files = [f for f in os.listdir(SKIN_DIR) if f.endswith(".png")]

def process_skin(filename):
path = os.path.join(SKIN_DIR, filename)
try:
img = Image.open(path).convert("RGB")
std = np.array(img).std()
if std >= STD_THRESHOLD:
shutil.copy2(path, os.path.join(OUTPUT_DIR, filename))
return "kept"
return "filtered"
except Exception:
return "error"

print(f"CPUs available: {cpu_count()}")

with Pool(processes=cpu_count()) as pool:
results = list(tqdm(
pool.imap(process_skin, skin_files, chunksize=100),
total=len(skin_files)
))

kept = results.count("kept")
filtered = results.count("filtered")
errors = results.count("error")

print(f"\nKept : {kept:,}")
print(f"Filtered : {filtered:,}")
print(f"Errors : {errors:,}")
```
then:
```python
import os
import zipfile
from tqdm import tqdm

INPUT_DIR = "/content/filtered_skins_plus"
ZIP_PATH = "/content/minecraft_skins_64x64_v1_1.zip"

skin_files = [f for f in os.listdir(INPUT_DIR) if f.endswith(".png")]
print(f"Skins to zip: {len(skin_files):,}")

with zipfile.ZipFile(ZIP_PATH, "w", zipfile.ZIP_DEFLATED, compresslevel=1) as zf:
for filename in tqdm(skin_files):
zf.write(os.path.join(INPUT_DIR, filename), arcname=filename)

size_mb = os.path.getsize(ZIP_PATH) / 1024 / 1024
print(f"\nDone!")
print(f"Skins zipped : {len(skin_files):,}")
print(f"ZIP size : {size_mb:.1f} MB")
```

任务类别：
- 文本到图像生成
- 图像分类
- 无条件图像生成
标签：
- Minecraft
- Minecraft皮肤
- 去重（de-duped）
- 去重（deduped）
- ZIP数据集
- ZIP
- ZIP归档（ZIP archive）
规模类别：
- 100万<样本数<1000万
许可证：Apache-2.0
友好名称：Minecraft皮肤110万去重版（64×64分辨率）1.1

# Minecraft皮肤110万去重版（64×64分辨率）1.1！
本数据集基于[Minecraft皮肤110万去重版](https://huggingface.co/datasets/MihaiPopa-1/minecraft-skins-1.1m-deduped-64x64)，但已过滤恶搞皮肤。

数据集格式为2.6GB的ZIP归档文件，内含64×64分辨率的PNG格式皮肤文件。

# 所用工具
Python图像库（PIL Image）及谷歌Colab（Google Colab）免费CPU算力版本。

# 数据集制作流程
1. 加载[Minecraft皮肤110万去重版](https://huggingface.co/datasets/MihaiPopa-1/minecraft-skins-1.1m-deduped-64x64)数据集；
2. 过滤恶搞皮肤（如单色皮肤），保留标准差（standard deviation，简称std）大于1的样本并移除其余内容；
3. 最终得到1,106,552张有效64×64分辨率皮肤；
4. 最终输出为2.6GB的ZIP归档文件。

本数据集可用于训练自定义皮肤生成模型（本人后续将采用矢量量化变分自编码器（VQ-VAE）模型进行开发）。

# 版本2的未来优化方向
1. 为皮肤添加文本标注（采用Florence 2基础版（Florence 2 Base）模型）；
2. ~~过滤恶搞皮肤（即仅由单一颜色构成的皮肤）~~ 已完成！

# 复现代码
用于复现该数据集的代码（全部由Claude 4.6 Sonnet生成）：
沿用此前代码，首先运行以下Python代码：
python
import os
import numpy as np
from PIL import Image
from tqdm import tqdm
from multiprocessing import Pool, cpu_count
import shutil

SKIN_DIR = "/content/filtered_skins"
OUTPUT_DIR = "/content/filtered_skins_plus"
STD_THRESHOLD = 1

os.makedirs(OUTPUT_DIR, exist_ok=True)
skin_files = [f for f in os.listdir(SKIN_DIR) if f.endswith(".png")]

def process_skin(filename):
path = os.path.join(SKIN_DIR, filename)
try:
img = Image.open(path).convert("RGB")
std = np.array(img).std()
if std >= STD_THRESHOLD:
shutil.copy2(path, os.path.join(OUTPUT_DIR, filename))
return "kept"
return "filtered"
except Exception:
return "error"

print(f"CPUs available: {cpu_count()}")

with Pool(processes=cpu_count()) as pool:
results = list(tqdm(
pool.imap(process_skin, skin_files, chunksize=100),
total=len(skin_files)
))

kept = results.count("kept")
filtered = results.count("filtered")
errors = results.count("error")

print(f"
Kept : {kept:,}")
print(f"Filtered : {filtered:,}")
print(f"Errors : {errors:,}")

随后运行以下代码以将处理后的皮肤打包为ZIP归档文件：
python
import os
import zipfile
from tqdm import tqdm

INPUT_DIR = "/content/filtered_skins_plus"
ZIP_PATH = "/content/minecraft_skins_64x64_v1_1.zip"

skin_files = [f for f in os.listdir(INPUT_DIR) if f.endswith(".png")]
print(f"Skins to zip: {len(skin_files):,}")

with zipfile.ZipFile(ZIP_PATH, "w", zipfile.ZIP_DEFLATED, compresslevel=1) as zf:
for filename in tqdm(skin_files):
zf.write(os.path.join(INPUT_DIR, filename), arcname=filename)

size_mb = os.path.getsize(ZIP_PATH) / 1024 / 1024
print(f"
Done!")
print(f"Skins zipped : {len(skin_files):,}")
print(f"ZIP size : {size_mb:.1f} MB")