new models

__init__.py

@@ -0,0 +1 @@
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.

glip.py

@@ -0,0 +1,120 @@
+import os
+import sys
+import numpy as np
+import torch
+import requests
+from time import time
+from io import BytesIO
+from pathlib import Path
+from PIL import Image
+
+# Optional plotting
+import matplotlib.pyplot as plt
+import matplotlib.pylab as pylab
+
+# Make sure GLIP is importable
+sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), "..")))
+
+from maskrcnn_benchmark.config import cfg
+from maskrcnn_benchmark.engine.predictor_glip import GLIPDemo
+
+# Constants
+GLIP_DIR = os.path.join(Path(__file__).parent.resolve(), "GLIP")
+WEIGHT_PATH = os.path.join(GLIP_DIR, "MODEL/glip_tiny_model_o365_goldg_cc_sbu.pth")
+CONFIG_FILE = os.path.join(GLIP_DIR, "configs/pretrain/glip_Swin_T_O365_GoldG.yaml")
+REMOTE_PATH = "https://penzhanwu2bbs.blob.core.windows.net/data/GLIPv1_Open/models/glip_tiny_model_o365_goldg_cc_sbu.pth"
+
+
+class GLIPPrediction:
+    def __init__(self, bboxes, scores, labels) -> None:
+        self.bboxes = bboxes
+        self.scores = scores
+        self.labels = labels
+        self.n = len(bboxes)
+
+
+class GLIP:
+    def __init__(self) -> None:
+        self._download_weight()
+        self._init_cfg(cfg)
+        self._check_gpu()
+        self._init_glip_model()
+
+    def _init_cfg(self, cfg):
+        self.cfg = cfg
+        self.cfg.local_rank = 0
+        self.cfg.num_gpus = 1
+        self.cfg.merge_from_file(CONFIG_FILE)
+        self.cfg.merge_from_list(["MODEL.WEIGHT", WEIGHT_PATH])
+        self.cfg.merge_from_list(["MODEL.DEVICE", "cuda"])
+
+    def _download_weight(self):
+        if not os.path.exists(WEIGHT_PATH):
+            Path(WEIGHT_PATH).parent.mkdir(parents=True, exist_ok=True)
+            os.system(f"wget {REMOTE_PATH} -O {WEIGHT_PATH}")
+
+    def _check_gpu(self):
+        try:
+            print("Check GPU info:")
+            print("CUDA available: {}".format(torch.cuda.is_available()))
+            print("CUDA version: {}".format(torch.version.cuda))
+            print("GPU count: {}".format(torch.cuda.device_count()))
+            print("GPU name: {}".format(torch.cuda.get_device_name(0)))
+        except Exception as e:
+            raise ValueError("Fail to check GPU info due to {}".format(e))
+
+    def _init_glip_model(self):
+        print("Initializing GLIP Model...")
+        start = time()
+        self.glip_model = GLIPDemo(
+            self.cfg,
+            min_image_size=800,
+            confidence_threshold=0.7,
+            show_mask_heatmaps=False,
+        )
+        print("GLIP Model Ready in {:.2f} seconds".format(time() - start))
+
+    def predict(self, image, query, debug=False) -> GLIPPrediction:
+        if len(image.shape) == 2:
+            image = np.stack([image] * 3, axis=-1)
+        predictions = self.glip_model.inference(image, query)
+        bboxes = predictions.bbox.tolist()
+        scores = predictions.get_field("scores").tolist()
+        labels = predictions.get_field("labels").tolist()
+        if debug:
+            print("bboxes: {}".format(bboxes))
+            print("scores: {}".format(scores))
+            print("labels: {}".format(labels))
+        class_names = query.split(" ")
+        classes = [class_names[label - 1] for label in labels]
+        return GLIPPrediction(bboxes, scores, classes)
+
+    def run_and_save(self, image, prompt, output_path="output.jpg"):
+        result_bgr, _ = self.glip_model.run_on_web_image(image, prompt, 0.5)
+        result_rgb = result_bgr[:, :, [2, 1, 0]]
+        result_image = Image.fromarray(result_rgb)
+        Path(output_path).parent.mkdir(parents=True, exist_ok=True)
+        result_image.save(output_path)
+        print(f"[✓] Saved result to: {output_path}")
+
+
+def load_image(source):
+    if source.startswith("http"):
+        response = requests.get(source)
+        pil_image = Image.open(BytesIO(response.content)).convert("RGB")
+    else:
+        pil_image = Image.open(source).convert("RGB")
+    image = np.array(pil_image)[:, :, [2, 1, 0]]
+    return image
+
+
+if __name__ == "__main__":
+    # Set your image source and prompt
+    image_path = "/home/vmukti/Pictures/car.jpg"  # or a URL
+    prompt = "white car in the leftmost side"
+    output_path = "outputs/output.jpeg"
+
+    # Run GLIP
+    glip = GLIP()
+    image = load_image(image_path)
+    glip.run_and_save(image, prompt, output_path)

maskrcnn_benchmark/modeling/__init__.py

@@ -0,0 +1 @@
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.