Merge pull request #4 from jangsoopark/v2.0.0

Noise simulation

Author: jangsoopark

README.md

@@ -146,7 +146,7 @@ $ python3 train.py
 ```
 
 #### Results of SOC
-- Final Accuracy is **99.18%**
+- Final Accuracy is **99.18%** (The official accuracy is 99.13%)
 - You can see the details in `notebook/experiments-SOC.ipynb`
 
 - Visualization of training loss and test accuracy
@@ -157,6 +157,15 @@ $ python3 train.py
 
 ![soc-confusion-matrix](./assets/figure/soc-confusion-matrix.png)
 
+- Noise Simulation [1]
+    - i.i.d samples from a uniform distribution
+
+| Noise | 1% | 5% | 10% | 15%|
+| :---: | :---: | :---: | :---: | :---: |
+| AConvNet-PyTorch | 98.56 | 94.39 | 85.03 | 73.65 |
+| AConvNet-Official | 91.76 | 88.52 | 75.84 | 54.68 |
+
+
 ### Extended Operating Conditions (EOC)
 
 ### Outlier Rejection
@@ -189,6 +198,12 @@ $ python3 train.py
 }
 ```
 
+## References
+[1] G. Dong, N. Wang, and G. Kuang, 
+"Sparse representation of monogenic signal: With application to target recognition in SAR images,"
+*IEEE Signal Process. Lett.*, vol. 21, no. 8, pp. 952-956, Aug. 2014.
+
+
 ---
 
 ## TODO

notebook/experiments-SOC.ipynb

@@ -138,14 +138,16 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 8,
+   "execution_count": 6,
    "metadata": {},
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "load test data set: 2425it [00:01, 2307.14it/s]\n"
+      "load test data set: 2425it [00:01, 2086.92it/s]\n",
+      "d:\\ivs\\project\\004-research\\signal-processing\\image-processing\\remote-sensing\\aconvnet\\aconvnet-pytorch\\venv\\lib\\site-packages\\torch\\nn\\functional.py:718: UserWarning: Named tensors and all their associated APIs are an experimental feature and subject to change. Please do not use them for anything important until they are released as stable. (Triggered internally at  ..\\c10/core/TensorImpl.h:1156.)\n",
+      "  return torch.max_pool2d(input, kernel_size, stride, padding, dilation, ceil_mode)\n"
      ]
     },
     {
@@ -212,7 +214,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 39,
+   "execution_count": 7,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -244,7 +246,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 48,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [
     {
@@ -277,6 +279,106 @@
     "plt.show()"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Noise Simulation"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 15,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from skimage import util"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 77,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "(614.0, 7744)"
+      ]
+     },
+     "execution_count": 77,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "util.random_noise(np.zeros((88, 88)), mode='s&p', amount=0.15).sum(), 88 * 88"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 99,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def generate_noise(_images, amount):\n",
+    "    \n",
+    "    n, _, h, w = _images.shape\n",
+    "        \n",
+    "    noise = np.random.uniform(size=(n, 1, h, w))\n",
+    "    portions = util.random_noise(np.zeros((n, 1, 88, 88)), mode='s&p', amount=amount)\n",
+    "    noise = noise * portions\n",
+    "    \n",
+    "    return _images + noise.astype(np.float32)\n",
+    "\n",
+    "\n",
+    "def noise_simulation(_m, ds, noise_ratio):\n",
+    "    \n",
+    "    num_data = 0\n",
+    "    corrects = 0\n",
+    "    \n",
+    "    _m.net.eval()\n",
+    "    _softmax = torch.nn.Softmax(dim=1)\n",
+    "    for i, data in enumerate(ds):\n",
+    "        images, labels = data\n",
+    "        images = generate_noise(images, noise_ratio)\n",
+    "\n",
+    "        predictions = _m.inference(images)\n",
+    "        predictions = _softmax(predictions)\n",
+    "\n",
+    "        _, predictions = torch.max(predictions.data, 1)\n",
+    "        labels = labels.type(torch.LongTensor)\n",
+    "        num_data += labels.size(0)\n",
+    "        corrects += (predictions == labels.to(m.device)).sum().item()\n",
+    "\n",
+    "    accuracy = 100 * corrects / num_data\n",
+    "    return accuracy"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 103,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "ratio = 0.01, accuracy = 98.56\n",
+      "ratio = 0.05, accuracy = 94.39\n",
+      "ratio = 0.10, accuracy = 85.03\n",
+      "ratio = 0.15, accuracy = 73.65\n"
+     ]
+    }
+   ],
+   "source": [
+    "noise_result = {}\n",
+    "\n",
+    "for ratio in [0.01, 0.05, 0.10, 0.15]:\n",
+    "    noise_result[ratio] = noise_simulation(m, test_set, ratio)\n",
+    "    print(f'ratio = {ratio:.2f}, accuracy = {noise_result[ratio]:.2f}')\n"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,