Switching to Airbus training dataset

.gitattributes

@@ -35,3 +35,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 demo/Pleiades_HD15_Miami_Marina.jpg filter=lfs diff=lfs merge=lfs -text
 demo/Satellite_Image_Marina_New_Zealand.jpg filter=lfs diff=lfs merge=lfs -text
+weights/best_mAP_epoch_20.pth filter=lfs diff=lfs merge=lfs -text
+demo/836f35381.jpg filter=lfs diff=lfs merge=lfs -text
+demo/848d2afef.jpg filter=lfs diff=lfs merge=lfs -text
+demo/82f13510a.jpg filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -1,4 +1,5 @@
 run_docker.sh
 **/.ipynb_checkpoints/
 **/__pycache__
-Makefile
+**/.DS_Store
+Makefile

README.md

@@ -2,7 +2,7 @@
 title: Ship Detection in Optical Satellite Imagery
 emoji: 🚢
 colorFrom: purple
-colorTo: yellow
+colorTo: blue
 sdk: docker
 pinned: false
 license: cc-by-nc-sa-4.0

app.py

@@ -23,13 +23,13 @@ MARGIN = OVERLAP / 2
 BATCH_SIZE = 16
 
 # CLASSES
-CLASSES = DOTADataset.CLASSES
+CLASSES = ['ship',]
 
 # Choose to use a config and initialize the detector
-config_file = 'oriented_rcnn_r50_fpn_1x_dota_le90.py'
+config_file = 'redet_re50_refpn_1x_dota_ms_rr_le90.py'
 
 # Setup a checkpoint file to load
-weights_file = 'weights/oriented_rcnn_r50_fpn_1x_dota_le90-6d2b2ce0.pth'
+weights_file = 'weights/best_mAP_epoch_20.pth'
 
 # check if GPU if available
 device = torch.device("cuda:0" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu")
@@ -60,12 +60,8 @@ def predict_image(img, threshold):
     end_time = time.time()
     #print(result)
 
-    # filter results
-    SELECTED = 6
-    result = [c if i == SELECTED else np.zeros((0, 6), dtype=np.float32) for i, c in enumerate(result)]
-
     # total number of predictions
-    infos = count = np.sum(result[SELECTED][:, -1] > threshold)
+    infos = np.sum(result[0][:, -1] > threshold)
 
     img_preds = model.show_result(img, result, score_thr=threshold, show=False)
     return img_preds, img.shape, infos, end_time - start_time
@@ -73,6 +69,9 @@ def predict_image(img, threshold):
 
 # Define example images and their true labels for users to choose from
 example_data = [
+    ["./demo/82f13510a.jpg", 0.75],
+    ["./demo/836f35381.jpg", 0.75],
+    ["./demo/848d2afef.jpg", 0.75],
     ["./demo/Satellite_Image_Marina_New_Zealand.jpg", 0.4],
     ["./demo/Pleiades_HD15_Miami_Marina.jpg", 0.4],
     # Add more example images and labels as needed
@@ -116,7 +115,7 @@ with demo:
         label='Try these images!'
     )
 
-    gr.Markdown("<p>This demo is provided by <a href='https://www.linkedin.com/in/faudi/'>Jeff Faudi</a> and <a href='https://www.dl4eo.com/'>DL4EO</a>. This model is based on the <a href='https://github.com/open-mmlab/mmrotate'>MMRotate framework</a> which provides oriented bounding boxes. We believe that oriented bouding boxes are better suited for detection in satellite images. This model has been trained on a combination of ships in VHR imagery datasets</p><p>The associated licenses are <a href='https://about.google/brand-resource-center/products-and-services/geo-guidelines/#google-earth-web-and-apps'>GoogleEarth fair use</a> and <a href='https://creativecommons.org/licenses/by-nc-sa/4.0/deed.en'>CC-BY-SA-NC</a>. This demonstration CANNOT be used for commercial puposes. Please contact <a href='mailto:jeff@dl4eo.com'>me</a> for more information on how you could get access to a commercial grade model or API. </p>")
+    gr.Markdown("<p>This demo is provided by <a href='https://www.linkedin.com/in/faudi/'>Jeff Faudi</a> and <a href='https://www.dl4eo.com/'>DL4EO</a>. This model is based on the <a href='https://github.com/open-mmlab/mmrotate'>MMRotate framework</a> which provides oriented bounding boxes. We believe that oriented bouding boxes are better suited for detection in satellite images. This model has been trained on Airbus Ship Detection available on Kaggle. The associated license is <a href='https://creativecommons.org/licenses/by-nc-sa/4.0/deed.en'>CC-BY-SA-NC</a>. This demonstration CANNOT be used for commercial puposes. Please contact <a href='mailto:jeff@dl4eo.com'>me</a> for more information on how you could get access to a commercial grade model or API. </p>")
 
 
 if os.path.exists('/.dockerenv'):
@@ -128,13 +127,13 @@ if os.path.exists('/.dockerenv'):
     demo.launch(
         server_name=hostname, 
         inline=False, 
-        #server_port=7860, 
+        server_port=7860, 
         debug=True
     )
 else:
     print('Not running inside a Docker container')
     demo.launch(
         inline=False, 
-        #server_port=7860, 
+        server_port=7860, 
         debug=False
     )

oriented_rcnn_r50_fpn_1x_dota_le90.py → redet_re50_refpn_1x_dota_ms_rr_le90.py

@@ -1,20 +1,28 @@
-dataset_type = 'DOTADataset'
-data_root = 'data/split_1024_dota1_0/'
+dataset_type = 'AirbusShipDataset'
+data_root = '/data/share/airbus-ship-detection/'
 img_norm_cfg = dict(
-    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+    mean=[52.29048625, 73.2539164, 80.97759001],
+    std=[53.09640994, 47.58987537, 42.15418378],
+    to_rgb=True)
 train_pipeline = [
     dict(type='LoadImageFromFile'),
     dict(type='LoadAnnotations', with_bbox=True),
-    dict(type='RResize', img_scale=(1024, 1024)),
+    dict(type='RResize', img_scale=(768, 768)),
     dict(
         type='RRandomFlip',
         flip_ratio=[0.25, 0.25, 0.25],
         direction=['horizontal', 'vertical', 'diagonal'],
         version='le90'),
+    dict(
+        type='PolyRandomRotate',
+        rotate_ratio=0.5,
+        angles_range=180,
+        auto_bound=False,
+        version='le90'),
     dict(
         type='Normalize',
-        mean=[123.675, 116.28, 103.53],
-        std=[58.395, 57.12, 57.375],
+        mean=[52.29048625, 73.2539164, 80.97759001],
+        std=[53.09640994, 47.58987537, 42.15418378],
         to_rgb=True),
     dict(type='Pad', size_divisor=32),
     dict(type='DefaultFormatBundle'),
@@ -24,14 +32,14 @@ test_pipeline = [
     dict(type='LoadImageFromFile'),
     dict(
         type='MultiScaleFlipAug',
-        img_scale=(1024, 1024),
+        img_scale=(768, 768),
         flip=False,
         transforms=[
-            dict(type='RResize'),
+            dict(type='RResize', img_scale=(768, 768)),
             dict(
                 type='Normalize',
-                mean=[123.675, 116.28, 103.53],
-                std=[58.395, 57.12, 57.375],
+                mean=[52.29048625, 73.2539164, 80.97759001],
+                std=[53.09640994, 47.58987537, 42.15418378],
                 to_rgb=True),
             dict(type='Pad', size_divisor=32),
             dict(type='DefaultFormatBundle'),
@@ -39,162 +47,209 @@ test_pipeline = [
         ])
 ]
 data = dict(
-    samples_per_gpu=2,
-    workers_per_gpu=2,
+    samples_per_gpu=20,
+    workers_per_gpu=8,
     train=dict(
-        type='DOTADataset',
-        ann_file='data/split_1024_dota1_0/trainval/annfiles/',
-        img_prefix='data/split_1024_dota1_0/trainval/images/',
+        type='AirbusShipDataset',
+        ann_file='full.csv',
+        img_prefix='train_v2/',
+        metrics_file='metrics_20240328.csv',
+        oversample_rate=5,
         pipeline=[
             dict(type='LoadImageFromFile'),
             dict(type='LoadAnnotations', with_bbox=True),
-            dict(type='RResize', img_scale=(1024, 1024)),
+            dict(type='RResize', img_scale=(768, 768)),
             dict(
                 type='RRandomFlip',
                 flip_ratio=[0.25, 0.25, 0.25],
                 direction=['horizontal', 'vertical', 'diagonal'],
                 version='le90'),
+            dict(
+                type='PolyRandomRotate',
+                rotate_ratio=0.5,
+                angles_range=180,
+                auto_bound=False,
+                version='le90'),
             dict(
                 type='Normalize',
-                mean=[123.675, 116.28, 103.53],
-                std=[58.395, 57.12, 57.375],
+                mean=[52.29048625, 73.2539164, 80.97759001],
+                std=[53.09640994, 47.58987537, 42.15418378],
                 to_rgb=True),
             dict(type='Pad', size_divisor=32),
             dict(type='DefaultFormatBundle'),
             dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
         ],
-        version='le90'),
+        version='le90',
+        data_root='/data/share/airbus-ship-detection/'),
     val=dict(
-        type='DOTADataset',
-        ann_file='data/split_1024_dota1_0/trainval/annfiles/',
-        img_prefix='data/split_1024_dota1_0/trainval/images/',
+        type='AirbusShipDataset',
+        ann_file='valid.csv',
+        img_prefix='train_v2/',
         pipeline=[
             dict(type='LoadImageFromFile'),
             dict(
                 type='MultiScaleFlipAug',
-                img_scale=(1024, 1024),
+                img_scale=(768, 768),
                 flip=False,
                 transforms=[
-                    dict(type='RResize'),
+                    dict(type='RResize', img_scale=(768, 768)),
                     dict(
                         type='Normalize',
-                        mean=[123.675, 116.28, 103.53],
-                        std=[58.395, 57.12, 57.375],
+                        mean=[52.29048625, 73.2539164, 80.97759001],
+                        std=[53.09640994, 47.58987537, 42.15418378],
                         to_rgb=True),
                     dict(type='Pad', size_divisor=32),
                     dict(type='DefaultFormatBundle'),
                     dict(type='Collect', keys=['img'])
                 ])
         ],
-        version='le90'),
+        version='le90',
+        data_root='/data/share/airbus-ship-detection/'),
     test=dict(
-        type='DOTADataset',
-        ann_file='data/split_1024_dota1_0/test/images/',
-        img_prefix='data/split_1024_dota1_0/test/images/',
+        type='AirbusShipDataset',
+        ann_file='valid.csv',
+        img_prefix='train_v2/',
         pipeline=[
             dict(type='LoadImageFromFile'),
             dict(
                 type='MultiScaleFlipAug',
-                img_scale=(1024, 1024),
+                img_scale=(768, 768),
                 flip=False,
                 transforms=[
-                    dict(type='RResize'),
+                    dict(type='RResize', img_scale=(768, 768)),
                     dict(
                         type='Normalize',
-                        mean=[123.675, 116.28, 103.53],
-                        std=[58.395, 57.12, 57.375],
+                        mean=[52.29048625, 73.2539164, 80.97759001],
+                        std=[53.09640994, 47.58987537, 42.15418378],
                         to_rgb=True),
                     dict(type='Pad', size_divisor=32),
                     dict(type='DefaultFormatBundle'),
                     dict(type='Collect', keys=['img'])
                 ])
         ],
-        version='le90'))
-evaluation = dict(interval=1, metric='mAP')
-optimizer = dict(type='SGD', lr=0.005, momentum=0.9, weight_decay=0.0001)
+        version='le90',
+        data_root='/data/share/airbus-ship-detection/'))
+evaluation = dict(interval=2, metric='mAP', save_best='mAP')
+optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001)
 optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
 lr_config = dict(
-    policy='step',
+    policy='CosineAnnealing',
     warmup='linear',
-    warmup_iters=500,
-    warmup_ratio=0.3333333333333333,
-    step=[8, 11])
-runner = dict(type='EpochBasedRunner', max_epochs=12)
-checkpoint_config = dict(interval=1)
-log_config = dict(interval=50, hooks=[dict(type='TextLoggerHook')])
+    warmup_iters=2000,
+    warmup_ratio=0.05,
+    min_lr_ratio=0.05)
+runner = dict(type='EpochBasedRunner', max_epochs=20)
+checkpoint_config = dict(interval=10)
+log_config = dict(
+    interval=200,
+    hooks=[dict(type='TextLoggerHook'),
+           dict(type='TensorboardLoggerHook')])
 dist_params = dict(backend='nccl')
 log_level = 'INFO'
-load_from = None
+load_from = 'redet_re50_fpn_1x_dota_ms_rr_le90-fc9217b5.pth'
 resume_from = None
 workflow = [('train', 1)]
 opencv_num_threads = 0
 mp_start_method = 'fork'
 angle_version = 'le90'
 model = dict(
-    type='OrientedRCNN',
+    type='ReDet',
     backbone=dict(
-        type='ResNet',
+        type='ReResNet',
         depth=50,
         num_stages=4,
         out_indices=(0, 1, 2, 3),
         frozen_stages=1,
-        norm_cfg=dict(type='BN', requires_grad=True),
-        norm_eval=True,
         style='pytorch',
-        init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+        pretrained='work_dirs/pretrain/re_resnet50_c8_batch256-25b16846.pth'),
     neck=dict(
-        type='FPN',
+        type='ReFPN',
         in_channels=[256, 512, 1024, 2048],
         out_channels=256,
         num_outs=5),
     rpn_head=dict(
-        type='OrientedRPNHead',
+        type='RotatedRPNHead',
         in_channels=256,
         feat_channels=256,
         version='le90',
         anchor_generator=dict(
             type='AnchorGenerator',
-            scales=[8],
-            ratios=[0.5, 1.0, 2.0],
+            scales=[2, 4],
+            ratios=[0.125, 0.5, 1.0, 2.0],
             strides=[4, 8, 16, 32, 64]),
         bbox_coder=dict(
-            type='MidpointOffsetCoder',
-            angle_range='le90',
-            target_means=[0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
-            target_stds=[1.0, 1.0, 1.0, 1.0, 0.5, 0.5]),
+            type='DeltaXYWHBBoxCoder',
+            target_means=[0.0, 0.0, 0.0, 0.0],
+            target_stds=[1.0, 1.0, 1.0, 1.0]),
         loss_cls=dict(
             type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
         loss_bbox=dict(
             type='SmoothL1Loss', beta=0.1111111111111111, loss_weight=1.0)),
     roi_head=dict(
-        type='OrientedStandardRoIHead',
-        bbox_roi_extractor=dict(
-            type='RotatedSingleRoIExtractor',
-            roi_layer=dict(
-                type='RoIAlignRotated',
-                out_size=7,
-                sample_num=2,
-                clockwise=True),
-            out_channels=256,
-            featmap_strides=[4, 8, 16, 32]),
-        bbox_head=dict(
-            type='RotatedShared2FCBBoxHead',
-            in_channels=256,
-            fc_out_channels=1024,
-            roi_feat_size=7,
-            num_classes=15,
-            bbox_coder=dict(
-                type='DeltaXYWHAOBBoxCoder',
-                angle_range='le90',
-                norm_factor=None,
-                edge_swap=True,
-                proj_xy=True,
-                target_means=(0.0, 0.0, 0.0, 0.0, 0.0),
-                target_stds=(0.1, 0.1, 0.2, 0.2, 0.1)),
-            reg_class_agnostic=True,
-            loss_cls=dict(
-                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
-            loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0))),
+        type='RoITransRoIHead',
+        version='le90',
+        num_stages=2,
+        stage_loss_weights=[1, 1],
+        bbox_roi_extractor=[
+            dict(
+                type='SingleRoIExtractor',
+                roi_layer=dict(
+                    type='RoIAlign', output_size=7, sampling_ratio=0),
+                out_channels=256,
+                featmap_strides=[4, 8, 16, 32]),
+            dict(
+                type='RotatedSingleRoIExtractor',
+                roi_layer=dict(
+                    type='RiRoIAlignRotated',
+                    out_size=7,
+                    num_samples=2,
+                    num_orientations=16,
+                    clockwise=True),
+                out_channels=256,
+                featmap_strides=[4, 8, 16, 32])
+        ],
+        bbox_head=[
+            dict(
+                type='RotatedShared2FCBBoxHead',
+                in_channels=256,
+                fc_out_channels=1024,
+                roi_feat_size=7,
+                num_classes=1,
+                bbox_coder=dict(
+                    type='DeltaXYWHAHBBoxCoder',
+                    angle_range='le90',
+                    norm_factor=2,
+                    edge_swap=True,
+                    target_means=[0.0, 0.0, 0.0, 0.0, 0.0],
+                    target_stds=[0.1, 0.1, 0.2, 0.2, 1]),
+                reg_class_agnostic=True,
+                loss_cls=dict(
+                    type='CrossEntropyLoss',
+                    use_sigmoid=False,
+                    loss_weight=1.0),
+                loss_bbox=dict(type='SmoothL1Loss', beta=1.0,
+                               loss_weight=1.0)),
+            dict(
+                type='RotatedShared2FCBBoxHead',
+                in_channels=256,
+                fc_out_channels=1024,
+                roi_feat_size=7,
+                num_classes=1,
+                bbox_coder=dict(
+                    type='DeltaXYWHAOBBoxCoder',
+                    angle_range='le90',
+                    norm_factor=None,
+                    edge_swap=True,
+                    proj_xy=True,
+                    target_means=[0.0, 0.0, 0.0, 0.0, 0.0],
+                    target_stds=[0.05, 0.05, 0.1, 0.1, 0.5]),
+                reg_class_agnostic=False,
+                loss_cls=dict(
+                    type='CrossEntropyLoss',
+                    use_sigmoid=False,
+                    loss_weight=1.0),
+                loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0))
+        ]),
     train_cfg=dict(
         rpn=dict(
             assigner=dict(
@@ -203,7 +258,8 @@ model = dict(
                 neg_iou_thr=0.3,
                 min_pos_iou=0.3,
                 match_low_quality=True,
-                ignore_iof_thr=-1),
+                ignore_iof_thr=-1,
+                gpu_assign_thr=200),
             sampler=dict(
                 type='RandomSampler',
                 num=256,
@@ -216,30 +272,49 @@ model = dict(
         rpn_proposal=dict(
             nms_pre=2000,
             max_per_img=2000,
-            nms=dict(type='nms', iou_threshold=0.8),
+            nms=dict(type='nms', iou_threshold=0.7),
             min_bbox_size=0),
-        rcnn=dict(
-            assigner=dict(
-                type='MaxIoUAssigner',
-                pos_iou_thr=0.5,
-                neg_iou_thr=0.5,
-                min_pos_iou=0.5,
-                match_low_quality=False,
-                iou_calculator=dict(type='RBboxOverlaps2D'),
-                ignore_iof_thr=-1),
-            sampler=dict(
-                type='RRandomSampler',
-                num=512,
-                pos_fraction=0.25,
-                neg_pos_ub=-1,
-                add_gt_as_proposals=True),
-            pos_weight=-1,
-            debug=False)),
+        rcnn=[
+            dict(
+                assigner=dict(
+                    type='MaxIoUAssigner',
+                    pos_iou_thr=0.5,
+                    neg_iou_thr=0.5,
+                    min_pos_iou=0.5,
+                    match_low_quality=False,
+                    ignore_iof_thr=-1,
+                    iou_calculator=dict(type='BboxOverlaps2D')),
+                sampler=dict(
+                    type='RandomSampler',
+                    num=512,
+                    pos_fraction=0.25,
+                    neg_pos_ub=-1,
+                    add_gt_as_proposals=True),
+                pos_weight=-1,
+                debug=False),
+            dict(
+                assigner=dict(
+                    type='MaxIoUAssigner',
+                    pos_iou_thr=0.5,
+                    neg_iou_thr=0.5,
+                    min_pos_iou=0.5,
+                    match_low_quality=False,
+                    ignore_iof_thr=-1,
+                    iou_calculator=dict(type='RBboxOverlaps2D')),
+                sampler=dict(
+                    type='RRandomSampler',
+                    num=512,
+                    pos_fraction=0.25,
+                    neg_pos_ub=-1,
+                    add_gt_as_proposals=True),
+                pos_weight=-1,
+                debug=False)
+        ]),
     test_cfg=dict(
         rpn=dict(
             nms_pre=2000,
             max_per_img=2000,
-            nms=dict(type='nms', iou_threshold=0.8),
+            nms=dict(type='nms', iou_threshold=0.7),
             min_bbox_size=0),
         rcnn=dict(
             nms_pre=2000,
@@ -247,3 +322,10 @@ model = dict(
             score_thr=0.05,
             nms=dict(iou_thr=0.1),
             max_per_img=2000)))
+img_size = 768
+max_keep_ckpts = 1
+val_dataloader = dict(samples_per_gpu=20, workers_per_gpu=8)
+seed = 1984
+gpu_ids = range(0, 1)
+device = 'cuda'
+work_dir = './logs/redet/2024-03-28-14-45-06'

weights/best_mAP_epoch_20.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31aebdccade8c5fc2ea4b81547a736a7a7648acc3c3fadc6337e46ff16943222
+size 363302067