Merge remote-tracking branch 'public/otx' into songkich/merge-dev-otx

models/segmentation/_base_/ocr_litehrnet18_mod2.py

@@ -1,7 +1,6 @@
 __norm_cfg = dict(type='BN', requires_grad=True)
 model = dict(
     type='ClassIncrSegmentor',
-    is_task_adapt=True,
     pretrained=None,
     backbone=dict(
         type='LiteHRNet',

models/segmentation/_base_/ocr_litehrnet_s_mod2.py

@@ -1,7 +1,6 @@
 __norm_cfg = dict(type='BN', requires_grad=True)
 model = dict(
     type='ClassIncrSegmentor',
-    is_task_adapt=True,
     pretrained=None,
     backbone=dict(
         type='LiteHRNet',

models/segmentation/_base_/ocr_litehrnet_x_mod3.py

@@ -1,7 +1,6 @@
 __norm_cfg = dict(type='BN', requires_grad=True)
 model = dict(
     type='ClassIncrSegmentor',
-    is_task_adapt=True,
     pretrained=None,
     backbone=dict(
         type='LiteHRNet',

mpa/cls/__init__.py

@@ -49,3 +49,5 @@
 import mpa.modules.models.losses.triplet_loss
 import mpa.modules.models.losses.barlowtwins_loss
 import mpa.modules.models.losses.mse_loss
+
+import mpa.modules.optimizer.lars

mpa/cls/explainer.py

@@ -14,12 +14,13 @@
 
 from mpa.registry import STAGES
 from mpa.cls.stage import ClsStage
-from mpa.modules.hooks.recording_forward_hooks import ActivationMapHook, EigenCamHook
+from mpa.modules.hooks.recording_forward_hooks import ActivationMapHook, EigenCamHook, ReciproCAMHook
 from mpa.utils.logger import get_logger
 logger = get_logger()
 EXPLAINER_HOOK_SELECTOR = {
     'eigencam': EigenCamHook,
     'activationmap': ActivationMapHook,
+    'classwisesaliencymap': ReciproCAMHook,
 }
 
 
@@ -35,9 +36,9 @@ def run(self, model_cfg, model_ckpt, data_cfg, **kwargs):
         explainer = kwargs.get('explainer')
         self.explainer_hook = EXPLAINER_HOOK_SELECTOR.get(explainer.lower(), None)
         if self.explainer_hook is None:
-            raise NotImplementedError(f'explainer algorithm {explainer} not supported')
+            raise NotImplementedError(f'Explainer algorithm {explainer} not supported!')
         logger.info(
-            f'explainer algorithm: {explainer}'
+            f'Explainer algorithm: {explainer}'
         )
         cfg = self.configure(model_cfg, model_ckpt, data_cfg, training=False, **kwargs)
 
@@ -71,7 +72,7 @@ def _explain(self, cfg):
             _ = load_checkpoint(model, cfg.load_from, map_location='cpu')
 
         model.eval()
-        model = MMDataParallel(model, device_ids=[0])
+        model = self._put_model_on_gpu(model, cfg)
 
         # InferenceProgressCallback (Time Monitor enable into Infer task)
         ClsStage.set_inference_progress_callback(model, cfg)

mpa/cls/inferrer.py

@@ -1,24 +1,21 @@
 # Copyright (C) 2022 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 #
+import os.path as osp
 from contextlib import nullcontext
 
-import os.path as osp
+import mmcv
 import numpy as np
 import torch
-
-import mmcv
-from mmcv.parallel import MMDataParallel
-from mmcv.runner import load_checkpoint, wrap_fp16_model
-
 from mmcls.datasets import build_dataloader, build_dataset
 from mmcls.models import build_classifier
-
-from mpa.registry import STAGES
+from mmcv.runner import load_checkpoint, wrap_fp16_model
 from mpa.cls.stage import ClsStage
-from mpa.modules.hooks.recording_forward_hooks import FeatureVectorHook, ReciproCAMHook
+from mpa.modules.hooks.recording_forward_hooks import ReciproCAMHook, FeatureVectorHook
 from mpa.modules.utils.task_adapt import prob_extractor
+from mpa.registry import STAGES
 from mpa.utils.logger import get_logger
+
 logger = get_logger()
 
 
@@ -83,7 +80,7 @@ def _infer(self, cfg, dump_features=False, dump_saliency_map=False):
             _ = load_checkpoint(model, cfg.load_from, map_location='cpu')
 
         model.eval()
-        model = MMDataParallel(model, device_ids=[0])
+        model = self._put_model_on_gpu(model, cfg)
 
         # InferenceProgressCallback (Time Monitor enable into Infer task)
         ClsStage.set_inference_progress_callback(model, cfg)
@@ -109,10 +106,9 @@ def _infer(self, cfg, dump_features=False, dump_saliency_map=False):
                     saliency_maps = forward_explainer_hook.records if dump_saliency_map else [None] * len(self.dataset)
 
         assert len(eval_predictions) == len(feature_vectors) == len(saliency_maps), \
-            (
-                "Number of elements should be the same, however, number of outputs are ",
-                f"{len(eval_predictions)}, {len(feature_vectors)}, and {len(saliency_maps)}"
-            )
+               "Number of elements should be the same, however, number of outputs are " \
+               f"{len(eval_predictions)}, {len(feature_vectors)}, and {len(saliency_maps)}"
+
         outputs = dict(
             eval_predictions=eval_predictions,
             feature_vectors=feature_vectors,

mpa/cls/stage.py

@@ -8,10 +8,12 @@
 
 from mmcv import ConfigDict
 from mmcv import build_from_cfg
+from mmcv.parallel import MMDataParallel, MMDistributedDataParallel
 
 from mpa.stage import Stage
 from mpa.utils.config_utils import update_or_add_custom_hook, recursively_update_cfg
 from mpa.utils.logger import get_logger
+from mpa.utils.data_cpu import MMDataCPU
 
 logger = get_logger()
 
@@ -118,7 +120,7 @@ def configure_model(cfg, training, **kwargs):
             return
 
         # update model layer's in/out configuration
-        from mmcls.models.builder import BACKBONES as backbone_reg
+        from mmcv.cnn import MODELS as backbone_reg
         layer = build_from_cfg(cfg.model.backbone, backbone_reg)
         layer.eval()
         input_shape = [3, 224, 224]
@@ -284,6 +286,26 @@ def configure_task(cfg, training, model_meta=None, **kwargs):
                 cfg.model.head.num_old_classes = len(old_classes)
         return model_tasks, dst_classes
 
+    def _put_model_on_gpu(self, model, cfg):
+        if torch.cuda.is_available():
+            model = model.cuda()
+            if self.distributed:
+                # put model on gpus
+                find_unused_parameters = cfg.get('find_unused_parameters', False)
+                # Sets the `find_unused_parameters` parameter in
+                # torch.nn.parallel.DistributedDataParallel
+                model = MMDistributedDataParallel(
+                    model,
+                    device_ids=[torch.cuda.current_device()],
+                    broadcast_buffers=False,
+                    find_unused_parameters=find_unused_parameters)
+            else:
+                model = MMDataParallel(
+                    model.cuda(), device_ids=[0])
+        else:
+            model = MMDataCPU(model)
+
+        return model
 
 def refine_tasks(train_cfg, meta, adapt_type):
     new_tasks = train_cfg['tasks']