-
|
I am having trouble loading a model that is shown here . If I use the RESNET load directly I have no problems. Debug shows the following lines: So several models are found with matching input and output. Documentation says the first one should be selected. Can anyone tell me what I am doing wrong? One thing I have noticed is the null GroupId of the matches is null. Also I have 2 engines loaded.Is this an issue? (Note: I have posted this in slack) |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments 5 replies
-
|
@hmf Can you provide your Criteria code? |
Beta Was this translation helpful? Give feedback.
-
|
@frankfliu I followed the example the only difference it being in Scala. Here is the code: import ai.djl.Application
import ai.djl.basicmodelzoo.BasicModelZoo
import ai.djl.modality.Classifications.Classification
import ai.djl.modality.cv.Image
import ai.djl.modality.cv.output.DetectedObjects
import ai.djl.repository.Artifact
import ai.djl.repository.zoo.{Criteria, ModelZoo, ZooModel}
import org.tensorflow.TensorFlow
import java.util
val criteria: Criteria[Image, Classification] = Criteria.builder()
.optApplication(Application.CV.IMAGE_CLASSIFICATION)
.setTypes(classOf[Image], classOf[Classification])
//.optFilter("layers", "50")
//.optFilter("flavor", "v1")
//.optFilter("dataset", "cifar10")
.build()
val ssd: ZooModel[Image, Classification] = ModelZoo.loadModel(criteria)I have noticed that not all criteria seem to have an effect on the selection. For example the val c0 = Criteria.builder()
.optModelZoo(new BasicModelZoo())
.optEngine("MXNet") // used, not working PyTorch, MXNet, Tensorflow
.optArtifactId("ssd") // set when application not use (#) yolo, coco, ssd
//.optGroupId("ai.djl.tensorflow") // ??
.optApplication(Application.CV.OBJECT_DETECTION) // used
.setTypes(classOf[Image], classOf[DetectedObjects])
.optFilter("dataset", "openimages_v4")
.build()
println(ModelZoo.listModels(c0))I get: Note that I would expect a single model due to the filter, but I don't. In addition to this if I add group id I nothing changes. What is even stranger is that if I add the code: println("000000000000000000")
val artifactId = c0.getArtifactId
val modelZoo: ModelZoo = c0.getModelZoo // Error not set
val groupId: String = c0.getGroupId
val engine: String = c0.getEngine
val application: Application = c0.getApplication
// Not used
val input = c0.getInputClass
val output = c0.getOutputClass
println(artifactId)
println(modelZoo)
println(groupId)
println(engine)
println(application)
println("** Not used")
println(input)
println(output)
println("11111111111111111111")I get: Why do I get back those nulls? TIA |
Beta Was this translation helpful? Give feedback.
-
|
Can you try switching your output type from |
Beta Was this translation helpful? Give feedback.
-
|
@zachgk That seems to solve it. Model reports the name val model: ZooModel[Image, Classifications] = BasicModelZoo.RESNET.loadModel(filter, ai.djl.Device.cpu(), pb)Still a newbie so what is the best way to figure out these types? EDIT: want to mark to as an answer but cannot seem to do it on your reply. |
Beta Was this translation helpful? Give feedback.
-
|
I am not sure if there is a good way at this point. If you want the easiest option, run a debugger and break somewhere in the Alternatively, you can inspect the code. Find the ModelZoo that you want and you will see a lot of ModelLoader static variables in it. Then, find the ModelLoader that you want from this list. Once you have the ModelLoader, check the constructor to see what TranslatorFactories are being added. Edit: |
Beta Was this translation helpful? Give feedback.
-
|
Should I report this as a documentation issue? |
Beta Was this translation helpful? Give feedback.
-
|
Sure. You are welcome to add issues or PRs if you notice places our documentation could be improved |
Beta Was this translation helpful? Give feedback.
-
|
reported here: #440 |
Beta Was this translation helpful? Give feedback.
@frankfliu I followed the example the only difference it being in Scala. Here is the code: