save_graph_format

[turian]

Is your feature request related to a problem? Please describe.

I am getting PNGs with graph_dir=LR that are too small to read.

Describe the solution you'd like I'd like save_graph_format so I could output "svg" or "PDF" or other formats that graphviz supports.

Describe alternatives you've considered Forking and hacking the format so it's not always PNG.

Screenshots / Text

My graph is 4839 x 325, which is just too small to read.

[mert-kurttutan]

Just so I understand correctly, Does the following code work for you? model_graph.visual_graph.render(format='svg')

If it does, I don't want to make draw_graph API too crowded.

Also, if preferred png (rather than svg) since it sometimes clips the image unexpectedly

[turian]

@mert-kurttutan I guess what I'm really looking for is a way to control the PNG height or width, so that it is not too small.

[mert-kurttutan]

There is the model_graph.resize_graph function whose one argument is scale. scale scales the image size. If you try scale > 1, the image should get bigger. You can try different values to see what value is high enough. At the moment, this is far from ideal essential because of how graphviz sets its image size, which depends many factors, node size, number of nodes etc.

Certainly, one thing that could be done is to use different default values of image size for horizontal setting

[turian]

This is great, thank you! Might I suggest this get added to the docs?

[turian]

There's apparently a lot more possible than I realized with this library :)

[mert-kurttutan]

This is great, thank you! Might I suggest this get added to the docs?

Sure. Any pr or suggestion for format would be valuable. I might change the format of docs. Do you think website dedicated to docs would be beneficial or is readme page enough for docs?

[kadirnar]

@mert-kurttutan

PNG

from torchyolo.predict import YoloHub

model = YoloHub(model_type="yolov5", model_path="yolov5n.pt", device="cuda:0", image_size=640)
model.view_model(file_format="png")

SVG

from torchyolo.predict import YoloHub

model = YoloHub(model_type="yolov5", model_path="yolov5n.pt", device="cuda:0", image_size=640)
model.view_model(file_format="svg")

[turian]

@mert-kurttutan I think README is fine, and a lot of people just look at what's on pypi.

On the other hand if you have good docstrings documenting the params, then turning that into documentation automatically could be useful.

[kvnptl]

model_graph.resize_graph(scale=5.0)
model_graph.visual_graph.render(format='svg')

Resizing and saving a graph as SVG is working fine for me.