the dataset of human.pt pre-weights

[ChenJian7578]

the dataset of human.pt from where?

[pderrenger]

@ChenJian7578 hello!

Thank you for your question. The human.pt pre-trained weights are typically trained on a variety of datasets to ensure robust performance across different scenarios. For the YOLO models, common datasets include COCO (Common Objects in Context), which contains a wide range of everyday objects, including humans.

If you are looking for specific details about the dataset used for a particular set of weights, you can often find this information in the model documentation or the release notes associated with the weights. For example, the COCO dataset is a popular choice for training object detection models and includes annotations for humans among other objects.

If you haven't already, I recommend checking the latest documentation and release notes for the most up-to-date information on the datasets used for training the pre-trained weights. Additionally, ensure you are using the latest version of the Ultralytics YOLO repository to benefit from the latest improvements and fixes.

If you have any more questions or need further assistance, feel free to ask! 😊

[soMarking]

I looked for the "ultralytics/ultralytics/nn/tasks.py" file, but I couldn't find the "HumanModel" or "humandetect" etc. in it.

always
AttributeError: Can't get attribute 'HumanModel' on <module 'ultralytics.nn.tasks' from 'C:\Users\HHD\Desktop\lifeVest\ultralytics\nn\tasks.py'>

[pderrenger]

@soMarking There’s no HumanModel in Ultralytics—only standard classes like DetectionModel/PoseModel/etc.—so that error means human.pt was saved in another repo that defined a custom class and can’t be unpickled by ultralytics.nn.tasks; load it in the original project or switch to an official YOLO11 person detector, for example:

from ultralytics import YOLO
model = YOLO("yolo11n.pt")
results = model("image.jpg", classes=[0])  # person

To inspect that file without its custom classes you can try:

from ultralytics.nn.tasks import torch_safe_load
ckpt, _ = torch_safe_load("human.pt", safe_only=True)

Please also confirm the issue persists on the latest ultralytics package; see the Reference for ultralytics.nn.tasks and torch_safe_load in the Ultralytics Docs for details. (docs.ultralytics.com)