#0:Add Trace Support

models/experimental/functional_UFLD_v2/README.md

@@ -1,22 +1,26 @@
 # Ultra-Fast-Lane-Detection-v2
 
-
+### Platforms:
+    WH N300
 
 ### Introduction
 
 The Ultra-Fast-Lane-Detection-v2 is a PyTorch-based implementation designed for fast and efficient deep lane detection using hybrid anchor-driven ordinal classification. It enhances the speed and accuracy of lane detection models with a focus on real-time applications.
 
-The repository includes models trained on various datasets such as [CULane](https://github.com/cfzd/Ultra-Fast-Lane-Detection-v2/blob/master/model/model_culane.py), [CurveLanes](https://github.com/cfzd/Ultra-Fast-Lane-Detection-v2/blob/master/model/model_curvelanes.py), and [Tusimple](https://github.com/cfzd/Ultra-Fast-Lane-Detection-v2/blob/master/model/model_tusimple.py)
-
 Resource link - [source](https://github.com/cfzd/Ultra-Fast-Lane-Detection-v2)
 
 Dataset Link - [source](https://github.com/TuSimple/tusimple-benchmark/issues/3), [kaggle](https://www.kaggle.com/datasets/manideep1108/tusimple?resource=download)
 
+
+The repository includes models trained on various datasets such as [CULane](https://github.com/cfzd/Ultra-Fast-Lane-Detection-v2/blob/master/model/model_culane.py), [CurveLanes](https://github.com/cfzd/Ultra-Fast-Lane-Detection-v2/blob/master/model/model_curvelanes.py), and [Tusimple](https://github.com/cfzd/Ultra-Fast-Lane-Detection-v2/blob/master/model/model_tusimple.py)
+
+
+
 ### Model Details
 
 - The entry point to the UFLD_v2 is located at:`models/experimental/functional_UFLD_v2/ttnn/ttnn_UFLD_v2.py`
 - The model picks up trained weights from the **tusimple_res34.pth** file located at:`models/experimental/functional_UFLD_v2/reference/tusimple_res34.pth`
-- Batch Size :1 (Currently supported)
+- Batch Size :2
 
 ### UFLD_v2 Demo
 
@@ -27,3 +31,9 @@ Dataset Link - [source](https://github.com/TuSimple/tusimple-benchmark/issues/3)
 - The output results for both torch and ttnn model will be stored in this directory(.txt files). - `models/experimental/functional_UFLD_v2/demo/results_txt`
 
 #### Note: The post-processing is performed using PyTorch.
+
+
+- Use the following command to run the UFLD_v2 performant implementation:
+  `pytest tests/ttnn/integration_tests/UFLD_v2/test_UFLD_v2_performant.py::test_run_ufld_v2_trace_2cq_inference`
+
+### Owner: [Venkatesh](https://github.com/vguduruTT)

models/experimental/functional_UFLD_v2/demo/demo.py

@@ -1,5 +1,16 @@
+# SPDX-FileCopyrightText: © 2025 Tenstorrent Inc.
+
+# SPDX-License-Identifier: Apache-2.0
+
 import torch
 import pytest
+import json
+import numpy as np
+import os
+import ttnn
+import gdown
+from pathlib import Path
+from loguru import logger
 from models.experimental.functional_UFLD_v2.reference.UFLD_v2_model import Tu_Simple
 from models.experimental.functional_UFLD_v2.demo import model_config as cfg
 from models.experimental.functional_UFLD_v2.demo.demo_utils import (
@@ -16,13 +27,6 @@
     preprocess_linear_weight,
     preprocess_linear_bias,
 )
-import json
-import numpy as np
-import os
-import ttnn
-import gdown
-from pathlib import Path
-from loguru import logger
 
 
 def custom_preprocessor(model, name):
@@ -294,29 +298,6 @@ def custom_preprocessor(model, name):
     return parameters
 
 
-def attempt_download(file, key="dumps"):
-    tests = Path(__file__).parent.parent / key
-    file_path = tests / Path(str(file).strip().replace("'", "").lower())
-    print("tests and file path is", tests, file_path)
-    if not file_path.exists():
-        google_drive_url = "https://drive.google.com/file/d/1pkz8homK433z39uStGK3ZWkDXrnBAMmX/view"
-
-        file_id = google_drive_url.split("/d/")[1].split("/")[0]
-        google_drive_download_url = f"https://drive.google.com/uc?id={file_id}"
-
-        try:
-            print(f"Downloading from Google Drive: {google_drive_download_url} to {file_path}...")
-            gdown.download(google_drive_download_url, str(file_path), quiet=False)
-
-            # Check if the file exists and its size
-            assert file_path.exists() and file_path.stat().st_size > 1e6, f"Download failed for {file}"
-
-        except Exception as e:
-            print(f"Error downloading from Google Drive: {e}")
-
-    return file_path
-
-
 @pytest.mark.parametrize(
     "batch_size,input_channels,height,width",
     [
@@ -327,11 +308,11 @@ def attempt_download(file, key="dumps"):
     "use_pretrained_weight",
     [
         # False,
-        True  # uncomment  to run the model for real weights
+        True
     ],
     ids=[
         # "pretrained_weight_false",
-        "pretrained_weight_true",  # uncomment to run the model for real weights
+        "pretrained_weight_true",
     ],
 )
 @pytest.mark.parametrize("device_params", [{"l1_small_size": 79104}], indirect=True)
@@ -341,8 +322,10 @@ def test_tu_simple_res34_inference(batch_size, input_channels, height, width, de
     if use_pretrained_weight:
         logger.info(f"Demo Inference using Pre-trained Weights")
         file = "tusimple_res34.pth"
-        downloaded_file_path = attempt_download(file, key="reference")
-        state_dict = torch.load(downloaded_file_path)
+        weights_path = "models/experimental/functional_UFLD_v2/tusimple_res34.pth"
+    if not os.path.exists(weights_path):
+        os.system("bash models/experimental/functional_UFLD_v2/weights_download.sh")
+        state_dict = torch.load(weights_path)
         new_state_dict = {}
         for key, value in state_dict["model"].items():
             new_key = key.replace("model.", "res_model.")

models/experimental/functional_UFLD_v2/demo/demo_utils.py

@@ -1,13 +1,15 @@
+# SPDX-FileCopyrightText: © 2025 Tenstorrent Inc.
+
+# SPDX-License-Identifier: Apache-2.0
+
+import ttnn
 import os, json
 import torch
 import torchvision.transforms as transforms
-from PIL import Image
 import numpy as np
+from PIL import Image
 from tqdm import tqdm
-import numpy as np
 from sklearn.linear_model import LinearRegression
-import json
-import ttnn
 
 
 def loader_func(path):
@@ -69,10 +71,7 @@ def bench(pred, gt, y_samples, running_time):
 
     @staticmethod
     def bench_one_submit(pred_file, gt_file):
-        # try:
         json_pred = [json.loads(line) for line in open(pred_file, "r").readlines()]
-        # except BaseException as e:
-        #     raise Exception('Fail to load json file of the prediction.')
         json_gt = [json.loads(line) for line in open(gt_file).readlines()]
         if len(json_gt) != len(json_pred):
             raise Exception("We do not get the predictions of all the test tasks")
@@ -122,7 +121,7 @@ def __init__(self, path, list_path, img_transform=None, crop_size=None):
         self.crop_size = crop_size
         with open(list_path, "r") as f:
             self.list = f.readlines()
-        self.list = [l[1:] if l[0] == "/" else l for l in self.list]  # exclude the incorrect path prefix '/' of CULane
+        self.list = [l[1:] if l[0] == "/" else l for l in self.list]
 
     def __getitem__(self, index):
         name = self.list[index].split()[0]
@@ -256,7 +255,6 @@ def generate_tusimple_lines(row_out, row_ext, col_out, col_ext, row_anchor=None,
             lanes_on_tusimple[tusimple_h_sample > bot_lim] = -2
             all_lanes.append(lanes_on_tusimple.tolist())
         else:
-            # all_lanes.append([-2]*56)
             pass
     return all_lanes
 
@@ -286,7 +284,7 @@ def run_test_tusimple(
         else:
             imgs = imgs.permute(0, 2, 3, 1)
             imgs = ttnn.from_torch(imgs, dtype=ttnn.bfloat16, layout=ttnn.ROW_MAJOR_LAYOUT, device=device)
-            out, pred = net(imgs)
+            out, pred = net(imgs, batch_size=batch_size)
             out = ttnn.to_torch(out)
             pred["loc_row"] = ttnn.to_torch(pred["loc_row"])
             pred["loc_col"] = ttnn.to_torch(pred["loc_col"])

models/experimental/functional_UFLD_v2/demo/model_config.py

@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: © 2025 Tenstorrent Inc.
+
+# SPDX-License-Identifier: Apache-2.0
+
 dataset = "Tusimple"
 data_root = "models/experimental/functional_UFLD_v2/demo"  # Need to be modified before running
 epoch = 100