Upload 6 files

README.md

@@ -0,0 +1,71 @@
+# Detectron2 Instance Segmentation Model
+
+This repository contains a Detectron2 model for instance segmentation. The model is a GeneralizedRCNN with a build_resnet_fpn_backbone backbone.
+
+## Model Details
+
+- **Architecture**: GeneralizedRCNN
+- **Backbone**: build_resnet_fpn_backbone
+- **Classes**: ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light', 'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear', 'hair drier', 'toothbrush']
+- **Training Dataset**: coco_2017_train
+
+## Usage with Detectron2
+
+```python
+import detectron2
+from detectron2.config import get_cfg
+from detectron2.modeling import build_model
+from detectron2.checkpoint import DetectionCheckpointer
+import torch
+import json
+
+# Set up configuration
+cfg = get_cfg()
+with open("config.json", "r") as f:
+    cfg_dict = json.load(f)
+    cfg.merge_from_dict(cfg_dict)
+
+# Build model
+model = build_model(cfg)
+
+# Load weights
+checkpointer = DetectionCheckpointer(model)
+checkpointer.load("model.pth")
+
+# Set model to evaluation mode
+model.eval()
+
+# For inference
+from detectron2.engine import DefaultPredictor
+predictor = DefaultPredictor(cfg)
+
+# Load an image
+import cv2
+image = cv2.imread("your_image.jpg")
+outputs = predictor(image)
+```
+
+## Sample Visualization Code
+
+```python
+from detectron2.utils.visualizer import Visualizer
+from detectron2.data import MetadataCatalog
+import cv2
+
+# Load class metadata
+with open("metadata.json", "r") as f:
+    metadata_dict = json.load(f)
+    
+# Create metadata
+metadata = MetadataCatalog.get("inference")
+metadata.thing_classes = metadata_dict["thing_classes"]
+
+# Visualize predictions
+v = Visualizer(image[:, :, ::-1], metadata=metadata, scale=1.2)
+out = v.draw_instance_predictions(outputs["instances"].to("cpu"))
+cv2.imwrite("output.jpg", out.get_image()[:, :, ::-1])
+```
+
+## Model Card for sajabdoli/detectron2-instance-segmentation
+
+This model is a Detectron2 implementation of instance segmentation. It can detect and segment objects in images.

config.json

@@ -0,0 +1 @@
+"CUDNN_BENCHMARK: false\nDATALOADER:\n  ASPECT_RATIO_GROUPING: true\n  FILTER_EMPTY_ANNOTATIONS: true\n  NUM_WORKERS: 4\n  REPEAT_SQRT: true\n  REPEAT_THRESHOLD: 0.0\n  SAMPLER_TRAIN: TrainingSampler\nDATASETS:\n  PRECOMPUTED_PROPOSAL_TOPK_TEST: 1000\n  PRECOMPUTED_PROPOSAL_TOPK_TRAIN: 2000\n  PROPOSAL_FILES_TEST: []\n  PROPOSAL_FILES_TRAIN: []\n  TEST:\n  - coco_2017_val\n  TRAIN:\n  - coco_2017_train\nFLOAT32_PRECISION: ''\nGLOBAL:\n  HACK: 1.0\nINPUT:\n  CROP:\n    ENABLED: false\n    SIZE:\n    - 0.9\n    - 0.9\n    TYPE: relative_range\n  FORMAT: BGR\n  MASK_FORMAT: polygon\n  MAX_SIZE_TEST: 1333\n  MAX_SIZE_TRAIN: 1333\n  MIN_SIZE_TEST: 800\n  MIN_SIZE_TRAIN:\n  - 640\n  - 672\n  - 704\n  - 736\n  - 768\n  - 800\n  MIN_SIZE_TRAIN_SAMPLING: choice\n  RANDOM_FLIP: horizontal\nMODEL:\n  ANCHOR_GENERATOR:\n    ANGLES:\n    - - -90\n      - 0\n      - 90\n    ASPECT_RATIOS:\n    - - 0.5\n      - 1.0\n      - 2.0\n    NAME: DefaultAnchorGenerator\n    OFFSET: 0.0\n    SIZES:\n    - - 32\n    - - 64\n    - - 128\n    - - 256\n    - - 512\n  BACKBONE:\n    FREEZE_AT: 2\n    NAME: build_resnet_fpn_backbone\n  DEVICE: cpu\n  FPN:\n    FUSE_TYPE: sum\n    IN_FEATURES:\n    - res2\n    - res3\n    - res4\n    - res5\n    NORM: ''\n    OUT_CHANNELS: 256\n  KEYPOINT_ON: false\n  LOAD_PROPOSALS: false\n  MASK_ON: true\n  META_ARCHITECTURE: GeneralizedRCNN\n  PANOPTIC_FPN:\n    COMBINE:\n      ENABLED: true\n      INSTANCES_CONFIDENCE_THRESH: 0.5\n      OVERLAP_THRESH: 0.5\n      STUFF_AREA_LIMIT: 4096\n    INSTANCE_LOSS_WEIGHT: 1.0\n  PIXEL_MEAN:\n  - 103.53\n  - 116.28\n  - 123.675\n  PIXEL_STD:\n  - 1.0\n  - 1.0\n  - 1.0\n  PROPOSAL_GENERATOR:\n    MIN_SIZE: 0\n    NAME: RPN\n  RESNETS:\n    DEFORM_MODULATED: false\n    DEFORM_NUM_GROUPS: 1\n    DEFORM_ON_PER_STAGE:\n    - false\n    - false\n    - false\n    - false\n    DEPTH: 50\n    NORM: FrozenBN\n    NUM_GROUPS: 1\n    OUT_FEATURES:\n    - res2\n    - res3\n    - res4\n    - res5\n    RES2_OUT_CHANNELS: 256\n    RES5_DILATION: 1\n    STEM_OUT_CHANNELS: 64\n    STRIDE_IN_1X1: true\n    WIDTH_PER_GROUP: 64\n  RETINANET:\n    BBOX_REG_LOSS_TYPE: smooth_l1\n    BBOX_REG_WEIGHTS: &id002\n    - 1.0\n    - 1.0\n    - 1.0\n    - 1.0\n    FOCAL_LOSS_ALPHA: 0.25\n    FOCAL_LOSS_GAMMA: 2.0\n    IN_FEATURES:\n    - p3\n    - p4\n    - p5\n    - p6\n    - p7\n    IOU_LABELS:\n    - 0\n    - -1\n    - 1\n    IOU_THRESHOLDS:\n    - 0.4\n    - 0.5\n    NMS_THRESH_TEST: 0.5\n    NORM: ''\n    NUM_CLASSES: 80\n    NUM_CONVS: 4\n    PRIOR_PROB: 0.01\n    SCORE_THRESH_TEST: 0.05\n    SMOOTH_L1_LOSS_BETA: 0.1\n    TOPK_CANDIDATES_TEST: 1000\n  ROI_BOX_CASCADE_HEAD:\n    BBOX_REG_WEIGHTS:\n    - &id001\n      - 10.0\n      - 10.0\n      - 5.0\n      - 5.0\n    - - 20.0\n      - 20.0\n      - 10.0\n      - 10.0\n    - - 30.0\n      - 30.0\n      - 15.0\n      - 15.0\n    IOUS:\n    - 0.5\n    - 0.6\n    - 0.7\n  ROI_BOX_HEAD:\n    BBOX_REG_LOSS_TYPE: smooth_l1\n    BBOX_REG_LOSS_WEIGHT: 1.0\n    BBOX_REG_WEIGHTS: *id001\n    CLS_AGNOSTIC_BBOX_REG: false\n    CONV_DIM: 256\n    FC_DIM: 1024\n    FED_LOSS_FREQ_WEIGHT_POWER: 0.5\n    FED_LOSS_NUM_CLASSES: 50\n    NAME: FastRCNNConvFCHead\n    NORM: ''\n    NUM_CONV: 0\n    NUM_FC: 2\n    POOLER_RESOLUTION: 7\n    POOLER_SAMPLING_RATIO: 0\n    POOLER_TYPE: ROIAlignV2\n    SMOOTH_L1_BETA: 0.0\n    TRAIN_ON_PRED_BOXES: false\n    USE_FED_LOSS: false\n    USE_SIGMOID_CE: false\n  ROI_HEADS:\n    BATCH_SIZE_PER_IMAGE: 512\n    IN_FEATURES:\n    - p2\n    - p3\n    - p4\n    - p5\n    IOU_LABELS:\n    - 0\n    - 1\n    IOU_THRESHOLDS:\n    - 0.5\n    NAME: StandardROIHeads\n    NMS_THRESH_TEST: 0.5\n    NUM_CLASSES: 80\n    POSITIVE_FRACTION: 0.25\n    PROPOSAL_APPEND_GT: true\n    SCORE_THRESH_TEST: 0.5\n  ROI_KEYPOINT_HEAD:\n    CONV_DIMS:\n    - 512\n    - 512\n    - 512\n    - 512\n    - 512\n    - 512\n    - 512\n    - 512\n    LOSS_WEIGHT: 1.0\n    MIN_KEYPOINTS_PER_IMAGE: 1\n    NAME: KRCNNConvDeconvUpsampleHead\n    NORMALIZE_LOSS_BY_VISIBLE_KEYPOINTS: true\n    NUM_KEYPOINTS: 17\n    POOLER_RESOLUTION: 14\n    POOLER_SAMPLING_RATIO: 0\n    POOLER_TYPE: ROIAlignV2\n  ROI_MASK_HEAD:\n    CLS_AGNOSTIC_MASK: false\n    CONV_DIM: 256\n    NAME: MaskRCNNConvUpsampleHead\n    NORM: ''\n    NUM_CONV: 4\n    POOLER_RESOLUTION: 14\n    POOLER_SAMPLING_RATIO: 0\n    POOLER_TYPE: ROIAlignV2\n  RPN:\n    BATCH_SIZE_PER_IMAGE: 256\n    BBOX_REG_LOSS_TYPE: smooth_l1\n    BBOX_REG_LOSS_WEIGHT: 1.0\n    BBOX_REG_WEIGHTS: *id002\n    BOUNDARY_THRESH: -1\n    CONV_DIMS:\n    - -1\n    HEAD_NAME: StandardRPNHead\n    IN_FEATURES:\n    - p2\n    - p3\n    - p4\n    - p5\n    - p6\n    IOU_LABELS:\n    - 0\n    - -1\n    - 1\n    IOU_THRESHOLDS:\n    - 0.3\n    - 0.7\n    LOSS_WEIGHT: 1.0\n    NMS_THRESH: 0.7\n    POSITIVE_FRACTION: 0.5\n    POST_NMS_TOPK_TEST: 1000\n    POST_NMS_TOPK_TRAIN: 1000\n    PRE_NMS_TOPK_TEST: 1000\n    PRE_NMS_TOPK_TRAIN: 2000\n    SMOOTH_L1_BETA: 0.0\n  SEM_SEG_HEAD:\n    COMMON_STRIDE: 4\n    CONVS_DIM: 128\n    IGNORE_VALUE: 255\n    IN_FEATURES:\n    - p2\n    - p3\n    - p4\n    - p5\n    LOSS_WEIGHT: 1.0\n    NAME: SemSegFPNHead\n    NORM: GN\n    NUM_CLASSES: 54\n  WEIGHTS: https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x/137849600/model_final_f10217.pkl\nOUTPUT_DIR: ./output\nSEED: -1\nSOLVER:\n  AMP:\n    ENABLED: false\n  BASE_LR: 0.02\n  BASE_LR_END: 0.0\n  BIAS_LR_FACTOR: 1.0\n  CHECKPOINT_PERIOD: 5000\n  CLIP_GRADIENTS:\n    CLIP_TYPE: value\n    CLIP_VALUE: 1.0\n    ENABLED: false\n    NORM_TYPE: 2.0\n  GAMMA: 0.1\n  IMS_PER_BATCH: 16\n  LR_SCHEDULER_NAME: WarmupMultiStepLR\n  MAX_ITER: 270000\n  MOMENTUM: 0.9\n  NESTEROV: false\n  NUM_DECAYS: 3\n  REFERENCE_WORLD_SIZE: 0\n  RESCALE_INTERVAL: false\n  STEPS:\n  - 210000\n  - 250000\n  WARMUP_FACTOR: 0.001\n  WARMUP_ITERS: 1000\n  WARMUP_METHOD: linear\n  WEIGHT_DECAY: 0.0001\n  WEIGHT_DECAY_BIAS: null\n  WEIGHT_DECAY_NORM: 0.0\nTEST:\n  AUG:\n    ENABLED: false\n    FLIP: true\n    MAX_SIZE: 4000\n    MIN_SIZES:\n    - 400\n    - 500\n    - 600\n    - 700\n    - 800\n    - 900\n    - 1000\n    - 1100\n    - 1200\n  DETECTIONS_PER_IMAGE: 100\n  EVAL_PERIOD: 0\n  EXPECTED_RESULTS: []\n  KEYPOINT_OKS_SIGMAS: []\n  PRECISE_BN:\n    ENABLED: false\n    NUM_ITER: 200\nVERSION: 2\nVIS_PERIOD: 0\n"

inference.py

@@ -0,0 +1,54 @@
+
+import torch
+import detectron2
+from detectron2.config import get_cfg
+from detectron2.engine import DefaultPredictor
+from detectron2.utils.visualizer import Visualizer
+from detectron2.data import MetadataCatalog
+import cv2
+import json
+import argparse
+
+def main():
+    parser = argparse.ArgumentParser(description="Run inference with Detectron2 model")
+    parser.add_argument("--image", required=True, help="Path to input image")
+    parser.add_argument("--output", default="output.jpg", help="Path to output image")
+    args = parser.parse_args()
+    
+    # Load config
+    cfg = get_cfg()
+    with open("config.json", "r") as f:
+        cfg_dict = json.load(f)
+        cfg.merge_from_dict(cfg_dict)
+    
+    # Update inference parameters
+    cfg.MODEL.WEIGHTS = "model.pth"
+    cfg.MODEL.DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+    
+    # Create predictor
+    predictor = DefaultPredictor(cfg)
+    
+    # Load image
+    image = cv2.imread(args.image)
+    
+    # Run prediction
+    outputs = predictor(image)
+    
+    # Load metadata
+    with open("metadata.json", "r") as f:
+        metadata_dict = json.load(f)
+    
+    # Setup metadata
+    metadata = MetadataCatalog.get("inference")
+    metadata.thing_classes = metadata_dict["thing_classes"]
+    
+    # Visualize
+    v = Visualizer(image[:, :, ::-1], metadata=metadata, scale=1.2)
+    out = v.draw_instance_predictions(outputs["instances"].to("cpu"))
+    
+    # Save output
+    cv2.imwrite(args.output, out.get_image()[:, :, ::-1])
+    print(f"Saved output to {args.output}")
+
+if __name__ == "__main__":
+    main()

metadata.json

@@ -0,0 +1 @@
+{"thing_classes": ["person", "bicycle", "car", "motorcycle", "airplane", "bus", "train", "truck", "boat", "traffic light", "fire hydrant", "stop sign", "parking meter", "bench", "bird", "cat", "dog", "horse", "sheep", "cow", "elephant", "bear", "zebra", "giraffe", "backpack", "umbrella", "handbag", "tie", "suitcase", "frisbee", "skis", "snowboard", "sports ball", "kite", "baseball bat", "baseball glove", "skateboard", "surfboard", "tennis racket", "bottle", "wine glass", "cup", "fork", "knife", "spoon", "bowl", "banana", "apple", "sandwich", "orange", "broccoli", "carrot", "hot dog", "pizza", "donut", "cake", "chair", "couch", "potted plant", "bed", "dining table", "toilet", "tv", "laptop", "mouse", "remote", "keyboard", "cell phone", "microwave", "oven", "toaster", "sink", "refrigerator", "book", "clock", "vase", "scissors", "teddy bear", "hair drier", "toothbrush"]}

model.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b990412f04e124ef3a22aad4604870357cbaa5453f10c090fd7666222448041
+size 177890206

requirements.txt

@@ -0,0 +1,5 @@
+
+torch>=1.7.0
+detectron2
+numpy
+opencv-python