README.md

---
library_name: diffusers
license: cc-by-4.0
pipeline_tag: unconditional-image-generation
---

# ddpm-spots-10-leopard

This model is part of my work for my [Freshman Seminar](https://www.teach.ustc.edu.cn/?attachment_id=17309) 
of my university.

It is an unconditional image generation model that outputs a $32\times 32$ grayscale image 
similar to those of 'leopard' category from the [SPOTS-10](https://github.com/Amotica/SPOTS-10) dataset.


## Uses

```py
from diffusers import DDPMPipeline
pipeline = DDPMPipeline.from_pretrained(
    "xinchengo/ddpm-spots-10-leopard",
).to("cuda")
image = pipeline().images[0]
image
```

## Training Details

### Training Data

the images labelled 'leopard' in the [SPOTS-10](https://github.com/Amotica/SPOTS-10) dataset

### Training Procedure

Mainly with [the procedure in the Diffusers tutorial](https://huggingface.co/docs/diffusers/tutorials/basic_training#train-a-diffusion-model) with a few modifications.


### Training Hyperparameters

```py
from dataclasses import dataclass

@dataclass
class TrainingConfig:
    image_size = 32
    train_batch_size = 64
    eval_batch_size = 16
    num_epochs = 50
    gradient_accumulation_steps = 1
    learning_rate = 1e-4
    lr_warmup_steps = 500
    save_image_epochs = 10
    save_model_epochs = 10
    mixed_precision = "fp16"
    output_dir = "ddpm-spots-10-leopard"

    push_to_hub = True
    hub_model_id = "xinchengo/ddpm-spots-10-leopard"
    hub_private_repo = None
    overwrite_output_dir = True
    seed = 0
```