422: "value is not a valid dict" on Form() BaseModel validation - @app.post uses Path(), File() and Form() - Request called from SwaggerUI

[baudneo]

First Check

• I added a very descriptive title to this issue.
• I used the GitHub search to find a similar issue and didn't find it.
• I searched the FastAPI documentation, with the integrated search.
• I already searched in Google "How to X in FastAPI" and didn't find any information.
• I already read and followed all the tutorial in the docs and didn't find an answer.
• I already checked if it is not related to FastAPI but to Pydantic.
• I already checked if it is not related to FastAPI but to Swagger UI.
• I already checked if it is not related to FastAPI but to ReDoc.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from enum import Enum

from fastapi import (
    FastAPI,
    File,
    UploadFile,
    Form,
)
from pydantic import BaseModel, Field

app = FastAPI()


# Models
class ModelProcessor(str, Enum):
    CPU = "cpu"
    GPU = "gpu"
    TPU = "tpu"
    DEFAULT = CPU


class ModelThresholds(BaseModel):
    confidence: float = Field(0.5, ge=0.0, le=1.0, descritpiton="Confidence Threshold")
    nms: float = Field(
        0.4, ge=0.0, le=1.0, description="Non-Maximum Suppression Threshold"
    )


class ModelOptions(BaseModel):
    processor: ModelProcessor = Field(None, description="Processor to use for model")
    height: int = Field(
        416, ge=1, description="Height of the input image (resized for model)"
    )
    width: int = Field(
        416, ge=1, description="Width of the input image (resized for model)"
    )
    square: bool = Field(False, description="Zero pad the image to be a square")
    fp_16: bool = Field(False, description="model uses Floating Point 16 Backend (EXPERIMENTAL!)")

    thresholds: ModelThresholds = Field(
        default_factory=ModelThresholds, description="Thresholds for the model"
    )


@app.post("/detect/{model_uuid}", summary="Run detection on an image")
async def object_detection(
        model_uuid: str,
        model_options: ModelOptions = Form(),
        image: UploadFile = File(..., description="Image to run the ML model on"),
):
    print(f"Running detection on image '{image.filename}'")
    return {"filename": image.filename}


if __name__ == "__main__":
    import uvicorn

    uvicorn.run(app, host="0.0.0.0", port=5001)

Description

What I am doing

• Open browser and head to http://localhost:5001/docs to open the SwaggerUI
• Try out the only defined endpoint, you can enter any string into model_uuid field and any image for the image: UploadFile, the model_options is prefilled with proper JSON based upon the BaseModel.
• Hit execute, a '422: Unprocessable Entity' error is returned.

What is happening

Error returned is 422: Unprocessable Entity ->

{
  "detail": [
    {
      "loc": [
        "body",
        "model_options"
      ],
      "msg": "value is not a valid dict",
      "type": "type_error.dict"
    }
  ]
}

What I expect to happen

I expect that the return statement is returned with a 200 response code.

Operating System

Linux

Operating System Details

No response

FastAPI Version

0.79.0

Python Version

3.10.6

Additional Context

[iudeen]

I don't think we can use a nested Pydantic model with Form() in a straightforward way. I will verify anyway

[sheilalarmour]

Hi all.

[mbroton]

Hi! As described in documentation:
Request Forms and Files

You can declare multiple File and Form parameters in a path operation, but you can't also declare Body fields that you expect to receive as JSON, as the request will have the body encoded using multipart/form-data instead of application/json.

This is not a limitation of FastAPI, it's part of the HTTP protocol.

The workaround for this would be to send JSON as a string and parse it on your own. But I'm not sure if it's recommended.

[baudneo]

Hi! As described in documentation:
https://fastapi.tiangolo.com/tutorial/request-forms-and-files/

You can declare multiple File and Form parameters in a path operation, but you can't also declare Body fields that you expect to receive as JSON, as the request will have the body encoded using multipart/form-data instead of application/json.

This is not a limitation of FastAPI, it's part of the HTTP protocol.

The workaround for this would be to send JSON as a string and parse it on your own. But I'm not sure if it's recommended.

Hi, I did see this in the docs but I was not certain of if that applied as in the same bit of docs it describes using Form() for JSON? Or maybe I am misunderstanding?

[iudeen]

The only limitation here is you will not be able to have nested JSONs within form.

from enum import Enum
from typing import Type
import inspect

from fastapi import (
    FastAPI,
    File,
    UploadFile,
    Form, Depends,
)
from pydantic import BaseModel, Field

app = FastAPI()

def as_form(cls: Type[BaseModel]):
    new_parameters = []

    for field_name, model_field in cls.__fields__.items():
        model_field: ModelField  # type: ignore

        new_parameters.append(
             inspect.Parameter(
                 model_field.alias,
                 inspect.Parameter.POSITIONAL_ONLY,
                 default=Form(...) if not model_field.required else Form(model_field.default),
                 annotation=model_field.outer_type_,
             )
         )

    async def as_form_func(**data):
        return cls(**data)

    sig = inspect.signature(as_form_func)
    sig = sig.replace(parameters=new_parameters)
    as_form_func.__signature__ = sig  # type: ignore
    setattr(cls, 'as_form', as_form_func)
    return cls

# Models
class ModelProcessor(str, Enum):
    CPU = "cpu"
    GPU = "gpu"
    TPU = "tpu"
    DEFAULT = CPU


class ModelThresholds(BaseModel):
    confidence: float = Field(0.5, ge=0.0, le=1.0, descritpiton="Confidence Threshold")
    nms: float = Field(
        0.4, ge=0.0, le=1.0, description="Non-Maximum Suppression Threshold"
    )

class ModelOptions(BaseModel):
    processor: ModelProcessor = Form(None, description="Processor to use for model")
    height: int = Form(
        416, ge=1, description="Height of the input image (resized for model)"
    )
    width: int = Form(
        416, ge=1, description="Width of the input image (resized for model)"
    )
    square: bool = Form(False, description="Zero pad the image to be a square")
    fp_16: bool = Form(False, description="model uses Floating Point 16 Backend (EXPERIMENTAL!)")

    # thresholds: ModelThresholds = Form(...)



@app.post("/detect/{model_uuid}", summary="Run detection on an image")
async def object_detection(
        model_uuid: str,
        model_options: ModelOptions = Depends(),
        image: UploadFile = File(..., description="Image to run the ML model on"),
):
    print(f"Running detection on image '{image.filename}'")
    return {"filename": image.filename}


if __name__ == "__main__":
    import uvicorn

    uvicorn.run(app, host="0.0.0.0", port=5001)

In the above code, I just commented out thresholds and changed how you have formed the BaseModel

[baudneo]

@iudeen - I am just following what the docs say about uploading an image with extra data -> https://fastapi.tiangolo.com/tutorial/request-forms-and-files/

Perhaps I am misunderstanding though?

[mbroton]

Examples in documentation have str type of Form. You've used ModelOptions.

You can use @iudeen's solution, but you can't nest your models.
Or send JSON as a string if you have to nest:

@app.post("/detect/{model_uuid}", summary="Run detection on an image")
async def object_detection(
        model_uuid: str,
        json_as_str: Json = Form(),  # from pydantic import Json
        image: UploadFile = File(..., description="Image to run the ML model on"),
):
    print(f"Running detection on image '{image.filename}'")
    return {"filename": image.filename}

which will result in this:

[baudneo]

Thank you for that, I am just trying to understand my error. So Form() will only accept str input? pydantic Json will take JSON formatted as a str and load it into a python dict?

Meaning I would need to instantiate a ModelOptions object using ModelOptions(**json_as_str) and then pass it to whatever function? My only concern thereafter would be the Swagger docs being incomplete on what the json_as_str should contain since it wont be able to reference ModelOptions.

Maybe I can set ModelOptions as Query()? So the end user knows what the options and defaults are?

Sorry if this seems obvious, I am just trying to understand.

[mbroton]

pydantic Json will take JSON formatted as a str and load it into a python dict?

Something like that.

I haven't tested it, but I guess you can do something like:

@app.post("/detect/{model_uuid}", summary="Run detection on an image")
async def object_detection(
        model_uuid: str,
        model_options: Json = Form(),
        image: UploadFile = File(..., description="Image to run the ML model on"),
):
    print(f"Running detection on image '{image.filename}'")
    options = ModelOptions.parse_obj(model_options)
    return {"filename": image.filename}

and options would be an instance of ModelOptions.

I can't provide answers for other questions, since I'm not sure.

[baudneo]

Thank you for the help, I am testing a few things out with the supplied options. Appreciate it very much @iudeen and @mbroton!

[iudeen]

@baudneo one last attempt to get as close as you want this to be. This handles everything.

import json
from enum import Enum
from typing import Type, Dict, Optional
import inspect

from fastapi import (
    FastAPI,
    File,
    UploadFile,
    Form, Depends,
)

from dataclasses import dataclass, field

from pydantic import BaseModel

app = FastAPI()


# Models
class ModelProcessor(str, Enum):
    CPU = "cpu"
    GPU = "gpu"
    TPU = "tpu"
    DEFAULT = CPU

class ModelThresholds(BaseModel):
    confidence: float = Form(0.5, ge=0.0, le=1.0, descritpiton="Confidence Threshold")
    nms: float = Form(
        0.4, ge=0.0, le=1.0, description="Non-Maximum Suppression Threshold"
    )

@dataclass
class ModelOptions:
    processor: ModelProcessor = Form(None, description="Processor to use for model")
    height: int = Form(
        416, ge=1, description="Height of the input image (resized for model)"
    )
    width: int = Form(
        416, ge=1, description="Width of the input image (resized for model)"
    )
    square: bool = Form(False, description="Zero pad the image to be a square")
    fp_16: bool = Form(False, description="model uses Floating Point 16 Backend (EXPERIMENTAL!)")

    thresholds: Optional[str] = Form(ModelThresholds().json())


@app.post("/detect/{model_uuid}", summary="Run detection on an image")
async def object_detection(
        model_uuid: str,
        model_options: ModelOptions = Depends(),
        image: UploadFile = File(..., description="Image to run the ML model on"),
):
    threshold = ModelThresholds(**json.loads(model_options.thresholds))
    print(threshold.dict())
    print(f"Running detection on image '{image.filename}'")
    return {"filename": image.filename}


if __name__ == "__main__":
    import uvicorn

    uvicorn.run(app, host="0.0.0.0", port=5001)