Added Ptv Transforms Resize node to resize image data. Added workflows and PyTorch model training tutorials.

2025-03-06 23:09:54 -08:00
parent 0d44bd51ec
commit 4f0fc89004
14 changed files with 340 additions and 58 deletions
@@ -0,0 +1,137 @@
+# Dog cat classification model training tutorial
+This tutorial covers how to use **ComfyUI Data Analysis** to train an image classification model with your own image dataset.  
+We'll use a **dog and cat image dataset** created specifically for this tutorial, which is available publicly on [my GitHub repo](https://github.com/HowToSD/dog_and_cat_dataset).  
+
+Once you complete the tutorial, you can experiment with your own images.
+
+**Disclaimer:** The model architecture is a traditional convolutional neural network (CNN) and is **not state-of-the-art**. This choice keeps the model small, enabling faster training and easier explanation.
+
+---
+
+## Overview of Steps
+
+1. Downloading the dog and cat dataset from the GitHub repo
+2. Edit the training workflow  
+3. Run the training workflow  
+4. Edit the evaluation workflow
+5. Run the evaluation workflow
+
+---
+
+## 1. Download the Dog & Cat Dataset  
+
+Visit [the GitHub repo](https://github.com/HowToSD/dog_and_cat_dataset) and go to the **Releases** page.  
+Click **"Source code (zip)"** to download the zip that contains all the images.  
+Extract the ZIP file using an unarchiver.  
+
+The dataset follows this directory structure:  
+
+```
+dog_and_cat
+  train
+    cats
+    dogs
+
+  val
+    cats
+    dogs
+```
+
+
+Take note of the **full paths** to the `train` and `val` directories, as you'll need them in the next step.
+
+---
+
+## 2. Editing the Training Workflow
+
+Drag and drop the below file into ComfyUI:  
+```
+examples/workflow/dog_cat_classification_train_model.json
+```
+Some nodes will be marked red, indicating they require edits.  
+The only required changes are the **train** and **val** dataset paths.  
+Replace them with the paths noted earlier.
+
+---
+
+## 3. Running the Training Workflow  
+
+Click **Queue** to start training.  
+
+Monitor the console for errors. If training runs smoothly, you should see output like this:
+
+```
+Epoch (train) 1/20, Loss: 0.6865
+Epoch (val) 1/20, Loss: 0.6801
+Epoch (train) 2/20, Loss: 0.6216
+Epoch (val) 2/20, Loss: 0.5710
+Epoch (train) 3/20, Loss: 0.4998
+Epoch (val) 3/20, Loss: 0.4716
+Epoch (train) 4/20, Loss: 0.3827
+Epoch (val) 4/20, Loss: 0.4045
+Epoch (train) 5/20, Loss: 0.2897
+Epoch (val) 5/20, Loss: 0.3297
+Epoch (train) 6/20, Loss: 0.2243
+Epoch (val) 6/20, Loss: 0.2925
+Epoch (train) 7/20, Loss: 0.1603
+Epoch (val) 7/20, Loss: 0.2916
+Epoch (train) 8/20, Loss: 0.1102
+Epoch (val) 8/20, Loss: 0.3216
+Epoch (train) 9/20, Loss: 0.0916
+Epoch (val) 9/20, Loss: 0.2706
+Epoch (train) 10/20, Loss: 0.0522
+Epoch (val) 10/20, Loss: 0.2220
+Epoch (train) 11/20, Loss: 0.0481
+Epoch (val) 11/20, Loss: 0.2820
+Epoch (train) 12/20, Loss: 0.0228
+Epoch (val) 12/20, Loss: 0.2899
+Epoch (train) 13/20, Loss: 0.0130
+Epoch (val) 13/20, Loss: 0.3056
+Epoch (train) 14/20, Loss: 0.0137
+Epoch (val) 14/20, Loss: 0.3070
+Epoch (train) 15/20, Loss: 0.0200
+Epoch (val) 15/20, Loss: 0.2714
+Epoch (train) 16/20, Loss: 0.0313
+Epoch (val) 16/20, Loss: 0.3108
+Epoch (train) 17/20, Loss: 0.0125
+Epoch (val) 17/20, Loss: 0.3388
+Epoch (train) 18/20, Loss: 0.0174
+Epoch (val) 18/20, Loss: 0.2989
+Epoch (train) 19/20, Loss: 0.0017
+Epoch (val) 19/20, Loss: 0.3165
+Epoch (train) 20/20, Loss: 0.0009
+Epoch (val) 20/20, Loss: 0.3582
+Prompt executed in 253.64 seconds
+```
+
+If you encounter an error:  
+- Ensure dataset paths are correct.  
+- Reduce **batch size** if your GPU runs out of memory (OOM error).  
+
+Once training completes, a **line chart** displaying training progress will appear at the bottom right of the canvas. You don’t need to interpret it initially.
+
+---
+
+## 4. Editing the Evaluation Workflow
+Now, let's verify if the model was trained correctly. In this step, we'll feed the validation data into the trained model and compute metrics to assess its performance.
+
+Drag and drop the below file into ComfyUI:  
+```
+examples/workflow/dog_cat_classification_eval_model.json
+```
+
+into **ComfyUI**.  
+
+Update the **val dataset path**.
+
+---
+
+## 5. Running the Evaluation Workflow  
+
+Click **Queue**. The evaluation process runs much faster than training.  
+
+Check the **Accuracy** node—it should display a value around **0.9**, confirming successful model training.
+
+Congratulations! You’ve successfully trained a machine learning model from scratch without writing a single line of code.
+
+Now, you can use your own images to train the model in the same way.
@@ -0,0 +1,49 @@
+# Model Training
+
+Experimental PyTorch model training has been added.  
+No coding is required. Training is as simple as a single click!  
+Currently, only classification is supported.
+
+The system is designed to be modular. For example, to switch from a linear model to a convolutional model, you only need to replace the model node (see screenshot below).
+
+You can choose to:
+* Use your own images
+* Use a public dataset
+
+## Using Your Own Images
+
+Check out [the dog and cat classification tutorial](dog_cat_classification_model_training.md).
+
+## Public Datasets
+
+Currently, only Fashion MNIST has been tested. Other public datasets have not been verified yet, though they may work.  
+If you try one and it doesn't work, please create an issue so I can investigate adding support.
+
+To use the Fashion MNIST dataset, simply drag and drop the workflow files below into ComfyUI. The model will be downloaded automatically.
+
+Linear model  
+* Training: examples/fashion_mnist_train.json  
+* Evaluation: examples/fashion_mnist_eval.json
+
+Convolutional model  
+* Training: examples/fashion_mnist_train_conv.json  
+* Evaluation: examples/fashion_mnist_eval_conv.json
+
+These training workflows train a FashionMNIST model from scratch effortlessly.
+
+### Training workflow
+**Linear model**
+![Train](images/fashion_mnist_train.png)
+
+**Conv model**
+![Train](images/conv_train.png)
+
+*Close up of the conv node*
+![Train](images/conv_train2.png)
+
+### Eval workflow
+**Linear model**
+![Eval](images/fashion_mnist_eval.png)
+
+**Conv model**
+![Eval](images/conv_eval.png)
@@ -276,6 +276,7 @@
 | [Ptv Dataset](ptv_dataset.md) | A Torchvision Dataset class wrapper. |
 | [Ptv Dataset Loader](ptv_dataset_loader.md) | A node to combine the dataset and data loader into a single node. |
 | [Ptv Image Folder Dataset](ptv_image_folder_dataset.md) | A Torchvision ImageFolder Dataset class wrapper. |
+| [Ptv Transforms Resize](ptv_transforms_resize.md) | Resizes and transforms elements of dataset to PyTorch tensors. |
 | [Ptv Transforms To Tensor](ptv_transforms_to_tensor.md) | Transforms elements of dataset to PyTorch tensors. |
 ## PyTorch wrapper - Transform
 | Node | Description |
@@ -0,0 +1,18 @@
+# Ptv Transforms Resize
+Resizes and transforms elements of dataset to PyTorch tensors.
+
+## Input
+| Name | Data type |
+|---|---|
+| height | Int |
+| width | Int |
+
+## Output
+| Data type |
+|---|
+| Ptvtransform |
+
+<HR>
+Category: PyTorch wrapper - Training
+
+ComfyUI Data Analysis Node Reference. © 2025 Hide Inada (HowToSD.com). All rights reserved.