Minor fixes related to infer and others (#1773)

Signed-off-by: Sachidanand Alle <[email protected]>

.dockerignore

@@ -14,6 +14,38 @@ __pycache__
 .vscode
 
 Dockerfile
-sample-apps/**/model
-node_modules
 .webpack
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+.venv/
+ubuntu_venv/
+
+# apps
+sample-apps/*/logs
+sample-apps/*/train
+sample-apps/*/model
+
+# docs build
+docs/build
+docs/source/apidocs
+
+# MONAI-label
+*.nii.gz
+.DS_Store
+*.DS_Store
+bin
+junit
+test-output.xml
+studies
+tests/data
+
+# Packages
+node_modules
+yarn.lock

.gitignore

@@ -120,6 +120,8 @@ venv/
 ENV/
 env.bak/
 venv.bak/
+.venv/
+ubuntu_venv/
 
 # Spyder project settings
 .spyderproject
@@ -139,12 +141,6 @@ dmypy.json
 # Pyre type checker
 .pyre/
 
-# virtualenv
-.venv/
-venv/
-ENV/
-ubuntu_venv/
-
 # IDE
 .idea/
 .vscode/
@@ -166,7 +162,7 @@ junit
 test-output.xml
 studies
 tests/data
-yarn.lock
+
 # Packages
 node_modules
 yarn.lock

monailabel/client/client.py

@@ -336,6 +336,7 @@ def infer(self, model, image_id, params, label_in=None, file=None, session_id=No
         fields = {"params": json.dumps(params) if params else "{}"}
         files = {"label": label_in} if label_in else {}
         files.update({"file": file} if file and not session_id else {})
+        logger.info(f"Files: {files}")
 
         status, form, files, _ = MONAILabelUtils.http_multipart(
             "POST", self._server_url, selector, fields, files, headers=self._headers
@@ -584,6 +585,9 @@ def send_response(conn, content_type="application/json"):
 
     @staticmethod
     def save_result(files, tmpdir):
+        if not files:
+            return None
+
         for name in files:
             data = files[name]
             result_file = os.path.join(tmpdir, name)

monailabel/endpoints/infer.py

@@ -91,10 +91,8 @@ def send_response(datastore, result, output, background_tasks):
     if output == "json":
         return res_json
 
-    m_type = get_mime_type(res_img)
-
     if output == "image":
-        return FileResponse(res_img, media_type=m_type, filename=os.path.basename(res_img))
+        return FileResponse(res_img, media_type=get_mime_type(res_img), filename=os.path.basename(res_img))
 
     if output == "dicom_seg":
         res_dicom_seg = result.get("dicom_seg")
@@ -106,7 +104,7 @@ def send_response(datastore, result, output, background_tasks):
     res_fields = dict()
     res_fields["params"] = (None, json.dumps(res_json), "application/json")
     if res_img and os.path.exists(res_img):
-        res_fields["image"] = (os.path.basename(res_img), open(res_img, "rb"), m_type)
+        res_fields["image"] = (os.path.basename(res_img), open(res_img, "rb"), get_mime_type(res_img))
     else:
         logger.info(f"Return only Result Json as Result Image is not available: {res_img}")
         return res_json

monailabel/interfaces/tasks/infer_v2.py

@@ -92,5 +92,5 @@ def is_valid(self) -> bool:
         pass
 
     @abstractmethod
-    def __call__(self, request) -> Union[Dict, Tuple[str, Dict[str, Any]]]:
+    def __call__(self, request) -> Tuple[Union[str, None], Dict]:
         pass

monailabel/tasks/infer/basic_infer.py

@@ -27,7 +27,7 @@
 from monailabel.interfaces.utils.transform import dump_data, run_transforms
 from monailabel.transform.cache import CacheTransformDatad
 from monailabel.transform.writer import ClassificationWriter, DetectionWriter, Writer
-from monailabel.utils.others.generic import device_list, device_map, name_to_device
+from monailabel.utils.others.generic import device_list, device_map, name_to_device, strtobool
 
 logger = logging.getLogger(__name__)
 
@@ -253,7 +253,7 @@ def detector(self, data=None) -> Optional[Callable]:
 
     def __call__(
         self, request, callbacks: Union[Dict[CallBackTypes, Any], None] = None
-    ) -> Union[Dict, Tuple[str, Dict[str, Any]]]:
+    ) -> Tuple[Union[str, None], Dict]:
         """
         It provides basic implementation to run the following in order
             - Run Pre Transforms
@@ -322,8 +322,8 @@ def __call__(
             data = callback_run_post_transforms(data)
         latency_post = time.time() - start
 
-        if self.skip_writer:
-            return dict(data)
+        if self.skip_writer or strtobool(data.get("skip_writer")):
+            return None, dict(data)
 
         start = time.time()
         result_file_name, result_json = self.writer(data)

monailabel/tasks/infer/bundle.py

@@ -137,15 +137,14 @@ def __init__(
 
         # labels = ({v.lower(): int(k) for k, v in pred.get("channel_def", {}).items() if v.lower() != "background"})
         labels = {}
+        type = self._get_type(os.path.basename(path), type)
         for k, v in pred.get("channel_def", {}).items():
-            if (not type.lower() == "deepedit") and (v.lower() != "background"):
-                labels[v.lower()] = int(k)
-            else:
+            logger.info(f"Model: {os.path.basename(path)}; Type: {type}; Label: {v} => {k}")
+            if v.lower() != "background" or type.lower() == "deepedit":
                 labels[v.lower()] = int(k)
         description = metadata.get("description")
         spatial_shape = image.get("spatial_shape")
         dimension = len(spatial_shape) if spatial_shape else 3
-        type = self._get_type(os.path.basename(path), type)
 
         # if detection task, set post restore to False by default.
         self.add_post_restore = False if type == "detection" else add_post_restore
@@ -273,28 +272,21 @@ def post_transforms(self, data=None) -> Sequence[Callable]:
         return post
 
     def _get_type(self, name, type):
+        if type:
+            return type
+
         name = name.lower() if name else ""
-        return (
-            (
-                InferType.DEEPEDIT
-                if "deepedit" in name
-                else (
-                    InferType.DEEPGROW
-                    if "deepgrow" in name
-                    else (
-                        InferType.DETECTION
-                        if "detection" in name
-                        else (
-                            InferType.SEGMENTATION
-                            if "segmentation" in name
-                            else InferType.CLASSIFICATION if "classification" in name else InferType.SEGMENTATION
-                        )
-                    )
-                )
-            )
-            if not type
-            else type
-        )
+        if "deepedit" in name:
+            return InferType.DEEPEDIT
+        if "deepgrow" in name:
+            return InferType.DEEPGROW
+        if "detection" in name:
+            return InferType.DETECTION
+        if "segmentation" in name:
+            return InferType.SEGMENTATION
+        if "classification" in name:
+            return InferType.CLASSIFICATION
+        return InferType.SEGMENTATION
 
     def _filter_transforms(self, transforms, filters):
         if not filters or not transforms:

monailabel/tasks/scoring/epistemic_v2.py

@@ -173,7 +173,7 @@ def run_scoring(self, image_id, simulation_size, model_ts, datastore):
 
         accum_unl_outputs = []
         for i in range(simulation_size):
-            data = self.infer_task(request=request)
+            _, data = self.infer_task(request=request)
             pred = data[self.infer_task.output_label_key] if isinstance(data, dict) else None
             if pred is not None:
                 logger.debug(f"EPISTEMIC:: {image_id} => {i} => pred: {pred.shape}; sum: {np.sum(pred)}")

monailabel/transform/post.py

@@ -129,7 +129,7 @@ def __call__(self, data):
             spatial_size = spatial_shape[-len(current_size) :]
 
             # Undo Spacing
-            if torch.any(torch.Tensor(np.not_equal(current_size, spatial_size))):
+            if np.any(np.not_equal(current_size, spatial_size)):
                 resizer = Resize(spatial_size=spatial_size, mode=self.mode[idx])
                 result = resizer(result, mode=self.mode[idx], align_corners=self.align_corners[idx])
 

monailabel/utils/others/generic.py

@@ -241,8 +241,8 @@ def _list_files(d, ext):
     ]
 
 
-def strtobool(str):
-    return bool(distutils.util.strtobool(str))
+def strtobool(s):
+    return False if s is None else s if isinstance(s, bool) else bool(distutils.util.strtobool(s))
 
 
 def is_openslide_supported(name):
@@ -336,7 +336,7 @@ def get_bundle_models(app_dir, conf, conf_key="models"):
     zoo_source = conf.get("zoo_source", settings.MONAI_ZOO_SOURCE)
 
     models = conf.get(conf_key)
-    models = models.split(",")
+    models = models.split(",") if models else []
     models = [m.strip() for m in models]
 
     if zoo_source == "monaihosting":  # if in github env, access model zoo

monailabel/utils/others/pathology.py

@@ -129,7 +129,7 @@ def create_asap_annotations_xml(json_data, loglevel="INFO"):
                 label = element["label"]
                 color = to_hex(color_map.get(label))
 
-                logger.info(f"Adding Contours for label: {label}; color: {color}; color_map: {color_map}")
+                logger.debug(f"Adding Contours for label: {label}; color: {color}; color_map: {color_map}")
                 labels[label] = color
 
                 contours = element["contours"]

sample-apps/endoscopy/lib/infers/deepedit.py

@@ -75,16 +75,21 @@ def inferer(self, data=None) -> Inferer:
         return SimpleInferer()
 
     def post_transforms(self, data=None) -> Sequence[Callable]:
-        return [
+        t = [
             EnsureTyped(keys="pred", device=data.get("device") if data else None),
             Activationsd(keys="pred", sigmoid=True),
             AsDiscreted(keys="pred", threshold=0.5),
             Restored(keys="pred", ref_image="image"),
             SqueezeDimd(keys="pred"),
             ToNumpyd(keys="pred", dtype=np.uint8),
-            FindContoursd(keys="pred", labels=self.labels, key_foreground_points="foreground"),
         ]
+        if data and data.get("output") == "mask":
+            return t
+        t.append(FindContoursd(keys="pred", labels=self.labels, key_foreground_points="foreground"))
+        return t
 
     def writer(self, data, extension=None, dtype=None):
+        if data and data.get("output") == "mask":
+            return super().writer(data, extension=extension, dtype=dtype)
         writer = PolygonWriter(label=self.output_label_key, json=self.output_json_key)
         return writer(data)

sample-apps/radiology/main.py

@@ -27,7 +27,6 @@
 from monailabel.interfaces.tasks.scoring import ScoringMethod
 from monailabel.interfaces.tasks.strategy import Strategy
 from monailabel.interfaces.tasks.train import TrainTask
-from monailabel.scribbles.infer import GMMBasedGraphCut, HistogramBasedGraphCut
 from monailabel.tasks.activelearning.first import First
 from monailabel.tasks.activelearning.random import Random
 
@@ -141,6 +140,8 @@ def init_infers(self) -> Dict[str, InferTask]:
         # Scribbles
         #################################################
         if self.scribbles:
+            from monailabel.scribbles.infer import GMMBasedGraphCut, HistogramBasedGraphCut
+
             infers.update(
                 {
                     "Histogram+GraphCut": HistogramBasedGraphCut(