# Filter4Free
**Repository Path**: fg_slash/filter4free
## Basic Information
- **Project Name**: Filter4Free
- **Description**: 使用神经网络拟合各种相机滤镜、胶片色彩
- **Primary Language**: Python
- **License**: Apache-2.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No
## Statistics
- **Stars**: 84
- **Forks**: 13
- **Created**: 2023-10-06
- **Last Updated**: 2025-10-18
## Categories & Tags
**Categories**: image-processing
**Tags**: None
## README
AI Film Simulation
Auther：Slash eMail：1364435561@qq.com
#### 1.Introduction
Use `neural networks` to fit various `camera filters` and `film colors`! The follow-up will `always update iteration` down, welcome to pay attention, welcome to the top right corner click `Started` ☆! This will be a great help, thank you! Geting for free is also welcome!
#### 2.Catalogue
- `【dir】static`：weight files and GUI resource files
- `checkpoints`：weight files
- `src`：resource files
- `【dir】idea`：theoretical research
- `src`：resource files
- `【dir】pack`：Nuitka package
- `【dir】script`：model convert script
- `train.py`：training script
- `train_new.py`：new training script
- `dataset.py`：dataset script
- `gui.py`：pyqt script
- `infer.py`：inference script
- `loss.py`：loss function
- `models.py`：models
- `preprocessing.py`：data preprocessing
- `utils`：utils
#### 3.Checkpoints
###### Negative color mask
* [x] `NegativeLabPro（NLP`）：**static/checkpoints/film-mask**
###### Olympus Film
* [x] `VIVID-浓郁色彩` ：**static/checkpoints/olympus/vivid**
* [ ] `SoftFocus-柔焦` ：
* [ ] `SoftLight-柔光` ：
* [ ] `Nostalgia-怀旧颗粒` ：
* [ ] `Stereoscopic-立体` ：
###### Fuji Film
* [x] `ACROS `：**static/checkpoints/fuji/acros**
* [x] `CLASSIC CHROME` ：**static/checkpoints/fuji/classic-chrome**
* [x] `ETERNA `：**static/checkpoints/fuji/enerna**
* [x] `ETERNA BLEACH BYPASS `：**static/checkpoints/fuji/eb**
* [x] `CLASSIC Neg. `：**static/checkpoints/fuji/classic-neg**
* [x] `PRO Neg.Hi `：**static/checkpoints/fuji/negstdhi**
* [x] `NOSTALGIC Neg.`：**static/checkpoints/fuji/nostalgic-neg**
* [x] `PRO Neg.Std` ：**static/checkpoints/fuji/negstd**
* [x] `ASTIA `：**static/checkpoints/fuji/astia**
* [x] `PROVIA`：**static/checkpoints/fuji/provia**
* [x] `VELVIA`：**static/checkpoints/fuji/velvia**
* [x] `Pro 400H`：**static/checkpoints/fuji/pro400h**
* [x] `Superia 400`：**static/checkpoints/fuji/superia400**
###### Kodak Film
* [x] `Color Plus`：**static/checkpoints/kodak/colorplus**
* [x] `Gold 200`：**static/checkpoints/kodak/gold200**
* [x] `Portra 400`：**static/checkpoints/kodak/portra400**
* [x] `Portra 160NC`：**static/checkpoints/kodak/portra160nc**
* [x] `UltraMax 400`：**static/checkpoints/kodak/ultramax400**
###### Rochi Film
* [ ] `Std-标准` ：
* [ ] `Vivid-鲜艳` ：
* [ ] `Single-单色` ：
* [ ] `SoftSingle-软单色` ：
* [ ] `StiffSingle-硬单色` ：
* [ ] `ContrastSingle-高对比对黑白` ：
* [ ] `Neg-负片` ：
* [ ] `R-Pos-正片` ：
* [ ] `R-Nostalgia-怀旧` ：
* [ ] `R-HDR-HDR` ：
* [ ] `R-Pos2Neg-正负逆冲` ：
###### Polaroid Film
* [x] `Polaroid`：**static/checkpoints/polaroid**
###### Sony Film
###### Nikon Film
###### Canon Film
###### Hasselblad Film
Model-Checkpoint
| Model | Checkpoint |
| ---------------- | ------------ |
| FilterSimulation | filmcnn.pth |
| UNet | unet.pth |
| UCM | best-ucm.pth |
**Notice: use unet model can get more details**
#### 4.Usage
###### Pycharm
1. Configuration environment：`pip install -r requirements.txt`.
2. Run `python gui.py` by GUI interface or run `python infer.py` by script inference.
###### Windows GUI
1. `Windows` Link
- 【V1.0】链接: https://pan.baidu.com/s/1WsBZbzCftyTMy3ZmzhJlDA 提取码: fmnq
- 【V1.1】链接: https://pan.baidu.com/s/1icLOXtVjUYqTkeDqf-o8Ag 提取码:e939
2. Run the `AIFilter.exe` executable in AIFilter.dist.
###### MacOS GUI
1. ` Silicon` Link (M1/M2/m3)`
- 【V1.0】链接: https://pan.baidu.com/s/1N5ux3eSUgYQTSB30iFw1GQ 提取码: nck8
- 【V1.1】链接: https://pan.baidu.com/s/1rnI5xPbwTkuZmetiWv0_6A 提取码: trbp
- 【V1.2】链接: https://pan.baidu.com/s/15v0pnFeGRMfCcVX5FE53_A 提取码: vp2x
2. `Intel` Link (i5/i7/i9)
- 【V1.0】链接: https://pan.baidu.com/s/14afbEXX_C4F7b-OeFHXRQg 提取码: mjc7
- 【V1.1】链接: https://pan.baidu.com/s/1SmBLFE7MT4KwxzbSzpJGYA 提取码: nbue
3. run `AIFilter.app` or move to `application`.
###### GUI(New design)
###### iPhone inference
convert pytorch to coreml，and infer on iphone14 pro, cost 4s when inferring a 12M image.
#### 5.Experimental Record
1.Refer to `idea/对比实验.md`, the framework is as follows:
2.Refer to `idea/自适应图像色彩迁移方案.md`, the framework is as follows:
#### 6.TODO
* [ ] Collect relevant image data to train more types of film simulations
* [x] Try to unify the mapping from all devices (iphone/ Android/Canon/Nikon, etc.) to film simulation
* [x] Porting to iPhone Inference
* [x] Supports onnx inference
* [x] Supports coreml inference
* [x] Supports TensorFlow inference
* [x] Supports mlx inference
#### 7.Update
1. 【24.01.21|Beta1.0】
- `Velvia`、`nn`、`nlp` 6 film filters.
- `GUI`
2. 【24.02.05|Beta1.1】
- Add 2 film filters : `nc`、`cc` .
- Add saving path prompt pop-up.
- Fix png image loading error, image format is opencv supported image type.
- Fix image loading display issue, retain original image scale for adaptive fill.
3. 【24.06.14】
- A new unified color simulation scheme is studied, which includes two stages of Decoloring and Coloring. For details, refer to `idea/自适应图像色彩迁移方案.md` .
4. 【24.07.4】
- Added provia filter, support for click/drag upload image, support for model switching.
5. 【24.09.20】
- Offer onnx、TensorFlow、mlx、coreml model converting script.
- Offer neural preset training script
6. 【25.01.31】
- Redesign the interface
- New Brand selection button
- Complete all film model training at Fuji
- Support raw formate：HEIF、RAF、CR2、CR3、RW2、DNG、PEF、ARW