256 to 512 resolution redaction

app.py

@@ -2,20 +2,23 @@ import gradio as gr
 
 import torch
 from fastai.vision.all import *
-from PIL import ImageFilter, ImageEnhance
+from PIL import ImageFilter, ImageEnhance, ImageDraw
 from diffusers.utils import make_image_grid
 from tqdm import tqdm
 from diffusers import AutoPipelineForInpainting, LCMScheduler, DDIMScheduler
 from diffusers import StableDiffusionControlNetInpaintPipeline, ControlNetModel
 import numpy as np
 from PIL import Image
+from datetime import datetime
 
 preferred_dtype = torch.float32
 preferred_device = "cuda" if torch.cuda.is_available() else "cpu"
 
 def label_func(fn): return path/"labels"/f"{fn.stem}_P{fn.suffix}"
 
-segmodel = load_learner("camvid-256.pkl")
+segmodel = load_learner("camvid-512.pkl")
+
+working_size = (512, 512)
 
 seg_vocabulary = ['Animal', 'Archway', 'Bicyclist', 'Bridge', 'Building', 'Car',
        'CartLuggagePram', 'Child', 'Column_Pole', 'Fence', 'LaneMkgsDriv',
@@ -59,9 +62,14 @@ def flip(img):
 
 
 def app(img):
-    npimg = np.array(img)
-    img = Image.fromarray(np.stack([npimg[:, :, 0], 255 - npimg[:,:,1], npimg[:,:,2]], axis=-1))
-    return img
+    start_time = datetime.now().timestamp()
+    img = img.resize(working_size)
+    mask = ban_cars_mask[get_seg_mask(img)]
+    overlay_img = Image.fromarray(np.stack([img[:, :, 0], mask / 2, img[:,:,2]], axis=-1))
+    end_time = datetime.now().timestamp()
+    draw = ImageDraw.Draw(overlay_img)
+    draw.text((10, 10), f"Duration: {int(1000 * (end_time - start_time))}ms", fill=(255, 255, 255))
+    return overlay_img
 
 #ideally:
 #iface = gr.Interface(app, gr.Image(sources=["webcam"], streaming=True), "image", live=True)

camvid-256.pkl → camvid-512.pkl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2049557e624452bcd5da711b90ea922756dde9aadccbefae6cfd360623fd2aff
-size 189310722
+oid sha256:18de31a47571cecd40053d46aa863a95365bfe70d67ebeffdfe4224bddb8ead7
+size 261871274