Updates and refactoring -- ready for production

pfms/controllers/pfmsController_inference.py

@@ -1,11 +1,13 @@
-from fastapi import APIRouter, Query, Request
-from fastapi import File, UploadFile
+from fastapi import APIRouter, Query, Request, UploadFile
+
 from fastapi.encoders import jsonable_encoder
 from fastapi.concurrency import run_in_threadpool
 from pydantic import BaseModel, Field
 from typing import Optional, List, Dict, Callable, Any
 
 import asyncio
+
+from torch import parse_ir
 from models import iresponse
 import os
 from datetime import datetime
@@ -26,6 +28,7 @@
 
 from spleenseg import spleenseg as spleen
 from spleenseg import splparser as psr
+from starlette.responses import FileResponse
 
 LOG = logger.debug
 
@@ -50,16 +53,39 @@ def noop():
     return {"status": True}
 
 
-def showAll() -> iresponse.modelsAvailable:
-    resp: iresponse.modelsAvailable = iresponse.modelsAvailable()
-    resp.models = ["spleensegmentation.pth"]
+def modelLocation_get(modelName) -> Path:
+    modelDir = settings.modelMeta.location / Path(modelName)
+    fileLocation: Path = modelDir / "model.pth"
+    return fileLocation
+
+
+async def model_save(
+    remote: UploadFile, modelName: str
+) -> iresponse.ModelUploadResponse:
+    modelUpload: iresponse.ModelUploadResponse = iresponse.ModelUploadResponse()
+    modelLocation: Path = modelLocation_get(modelName)
+    modelLocation.parent.mkdir(parents=True, exist_ok=True)
+    with modelLocation.open("wb") as buffer:
+        contents: bytes = await remote.read()
+        buffer.write(contents)
+    modelUpload.status = True
+    modelUpload.message = f"model {modelName} saved successfully"
+    modelUpload.location = modelLocation
+    return modelUpload
+
+
+def models_list() -> iresponse.ModelsAvailable:
+    resp: iresponse.ModelsAvailable = iresponse.ModelsAvailable()
+    resp.models = os.listdir(str(settings.modelMeta.location))
     return resp
 
 
 def IOpaths_create() -> tuple[Path, Path]:
     sessionUUID: UUID = uuid.uuid4()
-    baseParentPath: Path = Path(
-        Path(tempfile.mkdtemp()) / "imagesTs" / str(sessionUUID)
+    now: datetime = datetime.now()
+    nowstr: str = now.strftime("%Y-%m-%d-%H-%M-%S")
+    baseParentPath: Path = (
+        settings.analysisMeta.location / f"{nowstr}-{str(sessionUUID)}"
     )
     inputDir: Path = baseParentPath / "input"
     outputDir: Path = baseParentPath / "output"
@@ -68,25 +94,39 @@ def IOpaths_create() -> tuple[Path, Path]:
     return inputDir, outputDir
 
 
-def NIfTIvol_save(remote: File, inputDir: Path) -> Path:
-    savePath: Path = inputDir / "input.nii.gz"
+def NIfTIvol_saveInput(remote: UploadFile, inputDir: Path) -> Path:
+    saveDir: Path = inputDir / "imagesTs"
+    saveDir.mkdir(parents=True, exist_ok=True)
+    savePath: Path = saveDir / "input.nii.gz"
     with savePath.open("wb") as buffer:
         shutil.copyfileobj(remote.file, buffer)
     return savePath
 
 
-def NIfTIvol_infer(remote: File) -> Path:
-    inputDir: Path
-    outputDir: Path
-    inputDir, outputDir = IOpaths_create()
-    NIfTIvol_save(remote, inputDir)
-    options: Namespace = psr.parser_interpret(psr.parser_setup(""))
+def NIfTIvol_infer(remote: UploadFile, inputDir: Path, outputDir: Path) -> Path:
+    NIfTIvol_saveInput(remote, inputDir)
+    options: Namespace = psr.parser_interpret(psr.parser_setup(""), asModule=True)
     options.mode = "inference"
+    options.device = "cuda:0"
+    options.logTransformVols = True
+    options.inputdir = str(inputDir)
+    options.outputdir = str(outputDir)
     spleen.main(options, inputDir, outputDir)
-    return outputDir
+    outputFile: Path = outputDir / "inference" / "output.nii.gz"
+    return outputFile
+
 
+def model_prep(modelID: str, targetDir: Path):
+    modelLocation: Path = modelLocation_get(modelID)
+    shutil.copy(modelLocation, targetDir)
 
-async def inferenceOnNIfTI(uploadFile: File) -> iresponse.inferenceResponseNIFTI:
-    iresp: iresponse.inferenceResponseNIFTI = iresponse.inferenceResponseNIFTI()
-    result: Path = NIfTIvol_infer(uploadFile)
-    return iresp
+
+async def inferenceOnNIfTI(uploadFile: UploadFile, modelID: str = "") -> FileResponse:
+    inputDir: Path
+    outputDir: Path
+    inputDir, outputDir = IOpaths_create()
+    model_prep(modelID, inputDir)
+    result: Path = NIfTIvol_infer(uploadFile, inputDir, outputDir)
+    return FileResponse(
+        path=result, media_type="application/octet-stream", filename="output.nii.gz"
+    )