Fixed errors and added threshold configuration

config/datasets/anomaly_detection_images.yaml

@@ -22,4 +22,4 @@ test:
 
 infer:
   source: https://api.blackhole.ai.innopolis.university/public-datasets/anomaly_detection_mvtec/test.zip
-  target: ./data/MVTEC/test
+  target: ./data/MVTEC/infer

config/experiments/anomaly-detection/KG_210724_ba083ak_anomaly_detection_images.yaml

@@ -14,11 +14,5 @@ task: "anomaly-detection-images"
 random_seed: 0
 epochs: 50
 batch_size: 8
-accelerator: gpu
+ckpt_path: https://api.blackhole.ai.innopolis.university/pretrained/anomaly_det/anomaly50.pt
 
-wandb:
-  enable: True
-  project: anomaly_detect_mvtec
-  entity: "k-galliamov"
-  group: none
-  job_type: training

config/experiments/detection/KG_220524_coco.yaml

@@ -11,4 +11,4 @@ project: coco_pose_recognition
 random_seed: 0
 task: image-detection
 weights_freq: 1
-ckpt_path: https://api.blackhole.ai.innopolis.university/pretrained/pose_det/pose073.pt
+ckpt_path: https://api.blackhole.ai.innopolis.university/pretrained/pose_det/pose90.pt

config/experiments/example_image_classification_resnet.yaml

@@ -7,7 +7,6 @@ defaults:
   - override /augmentations_test: none
   - override /losses: log_loss.yaml
   - override /optimizers: adam
-  - override /schedulers:
 
 project: "mnist_classification"
 task: "image-classification"

config/experiments/semantic-segmentation/THRESHOLD.yaml

@@ -0,0 +1,18 @@
+# @package _global_
+defaults:
+  - override /models: semantic-segmentation/unet
+  - override /datasets: semantic-segmentation/segmentation_brain.yaml
+  - override /augmentations_train: none #segmentation_stroke.yaml
+  - override /augmentations_val: none
+  - override /augmentations_test: none
+  - override /losses: segmentation_losses.yaml
+models:
+  in_channels: 1
+project: "segmentation"
+task: "image-segmentation"
+random_seed: 42
+stop_param: 10
+epochs: 300
+weights_freq: 1
+batch_size: 10
+threshold: 0.2

config/models/detection/yolov5default.yaml

@@ -0,0 +1,3 @@
+_target_: ultralytics.YOLO
+description: yolov5 by ultralytics
+name: yolov5

dui/pages/config_edition.py

@@ -255,13 +255,16 @@ def on_strain_btn_button_click(set_progress, n, config_name):
                                 out_err = subprocess_errin.read()
 
                                 if out_err:
-                                    err_output.append(html.P(out_err))
+                                    out_err = out_err.split("\n")
+                                    for e in out_err:
+                                        err_output.append(html.P(e))
 
                                 if not out_text and process.poll() is None:
                                     time.sleep(0.5)
                                     continue
-
-                                text_output.append(html.P(out_text))
+                                out_text = out_text.split("\n")
+                                for t in out_text:
+                                    text_output.append(html.P(t))
 
                                 out = dbc.Row(id="process_output", children=[dbc.Col(text_output), dbc.Col(err_output)])
                                 set_progress(out)

infer.py

@@ -35,7 +35,7 @@ def main(config):
         ]
         config.experiment_name = experiment_name
     setup_clear_ml(config)
-    setup_wandb(config)
+    # setup_wandb(config)
     return run_pipeline(config, predict=True, test=False, train=False)
 
 

innofw/core/datamodules/lightning_datamodules/anomaly_detection_images.py

@@ -1,7 +1,7 @@
 import os
 import logging
 import pathlib
-
+from pathlib import Path
 import pandas as pd
 import torch
 import cv2
@@ -82,6 +82,8 @@ def predict_dataloader(self):
         return test_dataloader
 
     def setup_infer(self):
+        if str(self.predict_source).endswith("labels"):
+            self.predict_source = Path(str(self.predict_source)[:-6]+"images")
         self.predict_dataset = AnomaliesDataset(self.predict_source, self.get_aug(self.aug, 'test'))
 
     def save_preds(self, out_batches, stage: Stages, dst_path: pathlib.Path):

innofw/core/datamodules/lightning_datamodules/image_folder_dm.py

@@ -84,7 +84,7 @@ def setup_train_test_val(self, **kwargs):
         # divide into train, val, test
         n = len(train_dataset)
         train_size = int(n * (1 - self.val_size))
-
+        random_train_dataset, random_val_dataset = random_split(train_dataset, [train_size, n - train_size])
 
         # stratify
         if self.stratify:
@@ -98,11 +98,12 @@ def setup_train_test_val(self, **kwargs):
             strat_val = Counter(second_set_labels).values()
             coefsstrat_val = [i/len(second_set_labels) for i in strat_val]
 
-            random_train = Counter([self.train_dataset.dataset.targets[i] for i in self.train_dataset.indices]).values()
-            coefsrandom_train = [i/len(self.train_dataset.indices) for i in random_train]
 
-            random_val = Counter([self.val_dataset.dataset.targets[i] for i in self.val_dataset.indices]).values()
-            coefsrandom_val = [i/len(self.val_dataset.indices) for i in random_val]
+            random_train = Counter([random_train_dataset.dataset.targets[i] for i in random_train_dataset.indices]).values()
+            coefsrandom_train = [i/len(random_train_dataset.indices) for i in random_train]
+
+            random_val = Counter([random_val_dataset.dataset.targets[i] for i in random_val_dataset.indices]).values()
+            coefsrandom_val = [i/len(random_val_dataset.indices) for i in random_val]
 
             all_train = Counter(train_dataset.targets).values()
             coefsall_train = [i/len(train_dataset.targets) for i in all_train]
@@ -117,9 +118,9 @@ def setup_train_test_val(self, **kwargs):
             random train split - {coefsrandom_train}\n 
             random val split - {coefsrandom_val}\n ''')
         else:
-            self.train_dataset, self.val_dataset = random_split(
-                train_dataset, [train_size, n - train_size])
+            self.train_dataset, self.val_dataset = random_train_dataset, random_val_dataset
         # Set validatoin augmentations for val
+        setattr(self.train_dataset.dataset, "transform", train_aug)
         setattr(self.val_dataset.dataset, "transform", val_aug)
 
     def save_preds(self, preds, stage: Stages, dst_path: pathlib.Path):

innofw/core/models/torch/lightning_modules/anomaly_detection_images.py

@@ -99,13 +99,10 @@ def test_step(self, batch, batch_idx):
         return {"loss": loss}
 
     def predict_step(self, x, batch_idx, **kwargs):
-        return (x, self.compute_anomaly_mask(x))
-
-    def compute_anomaly_mask(self, x):
         x_rec = self.forward(x)  # (B, C, W, H)
         diff = ((x - x_rec) ** 2).sum(dim=1)  # sum across channels
-        mask = diff >= self.model.anomaly_threshold
-        return mask
+        mask = (diff >= self.model.anomaly_threshold).to(torch.uint8)
+        return x, mask
 
     def log_metrics(self, stage, metrics_res, *args, **kwargs):
         for key, value in metrics_res.items():

innofw/core/models/torch/lightning_modules/segmentation.py

@@ -1,4 +1,4 @@
-__all__ = ["SegmentationLM"]
+__all__ = ["SemanticSegmentationLightningModule", "MulticlassSemanticSegmentationLightningModule"]
 
 # standard libraries
 from typing import Any, Optional
@@ -177,7 +177,7 @@ def test_step(self, batch, *args, **kwargs) -> Optional[STEP_OUTPUT]:
     def model_load_checkpoint(self, path):
         self.model.load_state_dict(torch.load(path)["state_dict"])
 
-    def predict_step(self, batch: Any, batch_indx: int) -> torch.Tensor:
+    def predict_step(self, batch: Any, batch_indx: int, **kwargs) -> torch.Tensor:
         """Predict and output binary predictions"""
         if isinstance(batch, dict):
             input_tensor = batch[SegDataKeys.image]

innofw/core/models/torch_adapter.py

@@ -116,13 +116,22 @@ def __init__(
 
             if not value:
                 objects[key] = get_default(key, framework, task)
-
-        self.pl_module = objects["lightning_module"](
-            model,
-            objects["losses"],
-            objects["optimizers_cfg"],
-            objects["schedulers_cfg"],
-        )
+        if "threshold" in kwargs.keys() and kwargs["threshold"] is not None:
+            threshold = kwargs["threshold"]
+            self.pl_module = objects["lightning_module"](
+                model,
+                objects["losses"],
+                objects["optimizers_cfg"],
+                objects["schedulers_cfg"],
+                threshold=threshold
+            )
+        else:
+            self.pl_module = objects["lightning_module"](
+                model,
+                objects["losses"],
+                objects["optimizers_cfg"],
+                objects["schedulers_cfg"],
+            )
         try:
             self.pl_module.setup_metrics(self.metrics)
         except AttributeError:

innofw/pipeline.py

@@ -154,6 +154,7 @@ def run_pipeline(
         "weights_freq": cfg.get("weights_freq"),
         "logger": logger,
         "random_state": cfg.get("random_seed"),
+        "threshold": cfg.get("threshold")
     }
     inno_model = InnoModel(**model_params)
     result = None