No matching model with specified Input/Output type found.

Description

I’m trying to load a custom object detection model object detection model trained via a pytorch script. The model is saved as a .pt file. The mlnet.pt file is saved inside my eclipse workspace, which is where I’m running the code from. So the absolute path of the model to load is : “/Users/dpapp/eclipse-workspace/mlProject/src/mlnet/mlnet.pt”

I try to load this model using the following code:

Criteria<Image, DetectedObjects> criteria =
                Criteria.builder()
                        .optApplication(Application.CV.OBJECT_DETECTION)
                        .setTypes(Image.class, DetectedObjects.class)
                        .optModelUrls("file:///Users/dpapp/eclipse-workspace/mlProject/src/mlnet")
                        .build();
ZooModel model = criteria.loadModel();

However, I get the error message shown below.

Expected Behavior

I expect the model to successfully be loaded.

Error Message

ai.djl.repository.zoo.ModelNotFoundException: No matching model with specified Input/Output type found.
	at ai.djl.repository.zoo.Criteria.loadModel(Criteria.java:178)
	at main.main(main.java:86)
	at org.codehaus.mojo.exec.ExecJavaMojo$1.run(ExecJavaMojo.java:254)
	at java.base/java.lang.Thread.run(Thread.java:835)
Caused by: ai.djl.repository.zoo.ModelNotFoundException: No supported engine available for model zoo: ai.djl.localmodelzoo
	at ai.djl.repository.zoo.BaseModelLoader.loadModel(BaseModelLoader.java:135)
	at ai.djl.repository.zoo.Criteria.loadModel(Criteria.java:166)

How to Reproduce?

(If you developed your own code, please provide a short script that reproduces the error. For existing examples, please provide link.)

Steps to reproduce

(Paste the commands you ran that produced the error.)

1. Save mlnet.pt model in the following location: “/Users/dpapp/eclipse-workspace/mlProject/src/mlnet/mlnet.pt”
2. Criteria<Image, DetectedObjects> criteria = Criteria.builder() .optApplication(Application.CV.OBJECT_DETECTION) .setTypes(Image.class, DetectedObjects.class) .optModelUrls(“file:///Users/dpapp/eclipse-workspace/mlProject/src/mlnet”) .build(); ZooModel model = criteria.loadModel();

What have you tried to solve it?

1. I’ve tried both absolute and relative paths.
2. I’ve tried different criteria input and output types, including a generic <?, ?>
3. I’ve tried setting the optModelName parameter

Environment Info

maven inside eclipse

[frankfliu]

I went with the YoloTranslator and made some progress. However, I now get another error. This is what my translator looks like:

Translator<Image, DetectedObjects> translator = YoloTranslator.builder()
     .setPipeline(new Pipeline())
    .build();

And the error I’m getting is: ai.djl.engine.EngineException: PytorchStreamReader failed locating file constants.pkl: file not found at ai.djl.pytorch.jni.PyTorchLibrary.moduleLoad(Native Method) at ai.djl.pytorch.jni.JniUtils.loadModule(JniUtils.java:1360) at ai.djl.pytorch.engine.PtModel.load(PtModel.java:89)

My model only contains a .pt file. There are no other files, and definitely no constants.pkl file. However, when I was testing ObjectDetection using the tutorial: https://docs.djl.ai/jupyter/object_detection_with_model_zoo.html#step-1-load-image, I downloaded a model folder. That one had a constants.pkl file. Do you know what might be going on? Is my model not exported correctly? Thank you in advance!

You need use jit.trace() to export your model. libtorch C++ api only support traced script model: https://docs.djl.ai/docs/pytorch/how_to_convert_your_model_to_torchscript.html

See this PyTorch issue: https://github.com/pytorch/pytorch/issues/47917

[frankfliu]

@davpapp each SSD model processing are quite different, it’s really depends on how you trained it.

1. Synset, if your model is trained on Coco dataset, you can use the following:

    optSynsetUrl("https://mlrepo.djl.ai/model/cv/object_detection/ai/djl/pytorch/classes_coco.txt")

1. You can enable debug log level, you can see more model loading debug informations, see: https://github.com/deepjavalibrary/djl/blob/master/docs/load_model.md#enable-debug-log
2. The URL looks fine, but we recommend to use .optModelPath() if you load from local file system.