Add unit tests and PR description with issue link

Author: Samir-atra

official/projects/fcos/PR_DESCRIPTION.md

@@ -0,0 +1,41 @@
+# PR Description
+
+# Description
+
+Implementation of the **FCOS (Fully Convolutional One-Stage Object Detection)** model as a new project in the TensorFlow Model Garden. 
+
+This PR addresses issue #10275: **[Help wanted] FCOS: Fully Convolutional One-Stage Object Detection**.
+
+This works corresponds to the request to implement the [FCOS](https://arxiv.org/abs/1904.01355) paper. The implementation provides an anchor-free object detection framework with the following components:
+*   **Backbone**: ResNet50 (configurable via standard TF Model Garden mechanisms).
+*   **FPN**: Feature Pyramid Network for multi-scale feature extraction.
+*   **Heads**: Shared heads for Classification, Box Regression, and Centerness.
+*   **Losses**: IOULoss, FocalCrossEntropy, and BinaryCrossEntropy for centerness.
+
+## Type of change
+
+- [x] A new research paper code implementation
+- [x] New feature (non-breaking change which adds functionality)
+
+## Tests
+
+I have performed local testing to verify the integrity of the model architecture and the training loop.
+
+*   **Unit Tests**: Added `official/projects/fcos/model/model_test.py` to verify model instantiation and forward pass shapes.
+*   **Integration Test**: Ran `main.py` (training loop) locally with a subset of COCO data to verify the pipeline from data loading to gradient updates.
+
+**Test Configuration**:
+*   **Hardware**: Intel i7-1260P, NVIDIA RTX 3050 (Laptop environment)
+*   **OS**: Linux
+*   **TensorFlow Version**: 2.x
+
+## Checklist
+
+- [x] I have signed the [Contributor License Agreement](https://github.com/tensorflow/models/wiki/Contributor-License-Agreements). [User Action Required]
+- [x] I have read [guidelines for pull request](https://github.com/tensorflow/models/wiki/Submitting-a-pull-request).
+- [x] My code follows the [coding guidelines](https://github.com/tensorflow/models/wiki/Coding-guidelines).
+- [x] I have performed a self [code review](https://github.com/tensorflow/models/wiki/Code-review) of my own code.
+- [x] I have commented my code, particularly in hard-to-understand areas.
+- [x] I have made corresponding changes to the documentation (Added `README.md`).
+- [x] My changes generate no new warnings.
+- [x] I have added tests that prove my fix is effective or that my feature works.

official/projects/fcos/model/model_test.py

@@ -0,0 +1,38 @@
+
+import tensorflow as tf
+from absl.testing import parameterized
+from official.projects.fcos.model.model import FCOS
+
+class FCOSModelTest(parameterized.TestCase, tf.test.TestCase):
+
+    def test_fcos_model_instantiation_and_forward_pass(self):
+        """Tests that the FCOS model can be instantiated and run."""
+        
+        # Create the model
+        model = FCOS()
+        
+        # specific input shape [Batch, H, W, 3]
+        input_image_size = (1, 800, 1024, 3)
+        inputs = tf.random.normal(input_image_size)
+        
+        # Run forward pass
+        outputs = model(inputs, training=False)
+        
+        # Check output keys
+        self.assertIn('classifier', outputs)
+        self.assertIn('box', outputs)
+        self.assertIn('centerness', outputs)
+        
+        # Check output shapes (approximate checks based on feature map sizes)
+        # We expect a flattened output of shape [Batch, N_points, Channels]
+        self.assertEqual(outputs['classifier'].shape[0], 1)
+        self.assertEqual(outputs['box'].shape[0], 1)
+        self.assertEqual(outputs['centerness'].shape[0], 1)
+        
+        # Detailed shape check for channels
+        self.assertEqual(outputs['classifier'].shape[-1], 80) # 80 classes
+        self.assertEqual(outputs['box'].shape[-1], 4)       # 4 regression coords
+        self.assertEqual(outputs['centerness'].shape[-1], 1)# 1 centerness score
+
+if __name__ == '__main__':
+    tf.test.main()