Hello, i have trained a Gelan C-seg model for instance segmentation and i want to include it in my code to access the results directly to change the output image such as text font size. Due to large amount of images, i do not want to save images to my local workspace, i just want to display it and see if it is correct. For YOLOv8 i could do it with:
results = model.predict(), this gives me access to all output object such as orig_img, masks.xy etc. However, when i tried to run this by including Gelan C-seg yaml, the result output shows None for all masks. How do i resolve this? Thank you
It sounds like you’re on an exciting project with the Gelan C-seg model! If you’re seeing None for all masks, it might be due to a few reasons. Here are some steps you can take to troubleshoot and resolve the issue:
Check Model Compatibility: Ensure that the Gelan C-seg model is compatible with the Ultralytics framework you’re using. Sometimes, custom models may require specific configurations.
Verify YAML Configuration: Double-check your YAML configuration to ensure all paths and parameters are correctly set. Any discrepancies might lead to unexpected results.
Inspect Model Weights: Make sure the model weights are correctly loaded. You can try reloading the model to see if that resolves the issue.
Use Latest Version: Ensure you’re using the latest version of the Ultralytics package. Updates often include bug fixes and improvements.
Debugging: Add some print statements or use a debugger to inspect the intermediate outputs. This might help identify where things are going wrong.
Here’s a quick example of how you might structure your code to display results without saving:
from ultralytics import YOLO
# Load your model
model = YOLO('gelan-c-seg.yaml')
# Run prediction
results = model.predict('path/to/image.jpg')
# Access and display results
for result in results:
if result.masks is not None:
# Display or process your results here
result.show()
else:
print("No masks detected.")
When i ran another line i got another error:
model = YOLO(‘gelan-c-seg.yaml’)
File “C:\Users\User\AppData\Local\Programs\Python\Python311\Lib\site-packages\ultralytics\nn\tasks.py”, line 678, in parse_model
m = getattr(torch.nn, m[3:]) if ‘nn.’ in m else globals()[m] # get module
~~~~~~~~~^^^
KeyError: ‘RepNCSPELAN4’
From what i know, Gelan C-seg is YOLOv9, not YOLOv5 so i do not know where did the error generate from. However, please correct me if i’m wrong.
@Cereal if your best.pt model was not trained using the ultralytics library GitHub - ultralytics/ultralytics: Ultralytics YOLO11 🚀 and instead used another repository that was based on the YOLOv5 codebase (like the WongKinYiu repository), then it will not work as-is with ultralytics. You will need to train the model using the ultralytics package natively, as the weights files are not compatible with other code bases.
from ultralytics import YOLO
model = YOLO("yolov9c-seg.pt")
# Train model (if needed)
train_results = model.train(data="path/to/data.yaml") # update with other training arguments
model = YOLO("path/to/best.pt")
results = model.predict("path/to/source.png")
For brevity, I’ve included these all in a single block, but you’d likely want to separate out use of the predict() method to a new Python session. Once you have prediction results, you can see what methods are available to the Results object in the documentation Predict - Ultralytics YOLO Docs
You can test using the pretrained yolov9c-seg.pt model to ensure it works with the following:
from ultralytics import YOLO, ASSETS
model = YOLO("yolov9c-seg.pt")
results = model.predict(ASSETS / "bus.jpg")
for result in results:
print(result.masks.xy)
result.show() # preview image (does not wait, could open multiple windows for more than one file)
