Skip to content

Instantly share code, notes, and snippets.

@znewsham

znewsham/object_detection.ipynb

Created February 11, 2019 16:59
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save znewsham/ea6867005013715a4c43af3e901e06ec to your computer and use it in GitHub Desktop.
Save znewsham/ea6867005013715a4c43af3e901e06ec to your computer and use it in GitHub Desktop.
Download ZIP
Example of non-predictive object detection with fastai v1
Display the source blob
Display the rendered blob
Raw
object_detection.ipynb
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Image segmentation with CamVid"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"%reload_ext autoreload\n",
"%autoreload 2\n",
"%matplotlib inline"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"from fastai.vision import *\n",
"from fastai.callbacks.hooks import *\n",
"from sklearn.metrics import jaccard_similarity_score as jsc"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[PosixPath('/home/ubuntu/.fastai/data/coco_sample/train_sample'),\n",
" PosixPath('/home/ubuntu/.fastai/data/coco_sample/annotations')]"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"path = untar_data(URLs.COCO_SAMPLE)\n",
"path.ls()\n"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"path_train_sample = path/'train_sample/'\n",
"#path_annotations = path/'annotations'\n",
"#print(path_annotations.ls())\n",
"images, lbl_bbox = get_annotations(path/'annotations/train_sample.json')\n",
"img2bbox = dict(zip(images, lbl_bbox))\n",
"\n",
"\n",
"\n",
"classes=[\"chair\",\"couch\", \"tv\", \"remote\", \"book\", \"vase\"];\n",
"def get_lrg(b):\n",
" if not b: raise Exception()\n",
" b = sorted(b, key=lambda x: np.product(np.array(x[-2:])-np.array(x[:2])), \n",
" reverse=True)\n",
" return b[0]\n",
"trn_lrg_anno = {a: get_lrg(img2bbox[a][0]) for a in images}\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Data"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"fnames = get_image_files(path_train_sample)\n",
"fnames[:3]\n",
"img=open(fnames[0])"
]
},
{
"cell_type": "code",
"execution_count": 28,
"metadata": {},
"outputs": [],
"source": [
"def get_y_func(o):\n",
" #return FloatList(img2bbox[o.name][0][0])\n",
" points1 = trn_lrg_anno[o.name];\n",
" points = [\n",
" [points1[0], points1[1]],\n",
" [points1[2], points1[3]],\n",
" #[points1[2], points1[1]],\n",
" #[points1[0], points1[3]]\n",
" ]\n",
" #labels = list(map(lambda x: [1., 1.] if img2bbox[o.name][1][0] == x else [0., 0.], classes))\n",
" return tensor(points).float();\n",
"\n",
"#get_y_func(path_train_sample/\"000000066154.jpg\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Datasets"
]
},
{
"cell_type": "code",
"execution_count": 29,
"metadata": {
"scrolled": true
},
"outputs": [
{
"data": {
"text/plain": [
"21838"
]
},
"execution_count": 29,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"size = 224\n",
"bs=16\n",
"len(path_train_sample.ls())"
]
},
{
"cell_type": "code",
"execution_count": 37,
"metadata": {},
"outputs": [],
"source": [
"np.random.seed(2)\n",
"data = (PointsItemList.from_folder(path_train_sample)\n",
" .random_split_by_pct(0.95)\n",
" .label_from_func(get_y_func)\n",
" .transform(get_transforms(max_zoom=0., max_rotate=3.),tfm_y=True, resize_method=ResizeMethod.SQUISH, size=size, remove_out=False)\n",
" )\n",
"\n",
"data.lists[1].y = data.lists[1].y[0:1000]\n",
"data.lists[1].x = data.lists[1].x[0:1000]\n",
"data = data.databunch(bs=64).normalize(imagenet_stats)\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"data.show_batch(2, figsize=(10,7))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"data.show_batch(2, figsize=(10,7), ds_type=DatasetType.Valid)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Model"
]
},
{
"cell_type": "code",
"execution_count": 41,
"metadata": {},
"outputs": [],
"source": [
"head_reg4 = nn.Sequential(Flatten(), nn.Linear(25088,4))\n",
"lmse=MSELossFlat();\n",
"def accuracy1(input, targs):\n",
" input = input.view(-1)\n",
" targs = targs.view(-1)\n",
" return lmse((input + 1) * 112, (targs + 1) * 112).sqrt();\n",
" \n",
"l1loss = nn.L1Loss();\n",
"def loss_func(input, targs):\n",
" #print(input.view(-1).shape, targs.view(-1).shape);\n",
" return l1loss(input.view(-1), targs.view(-1));\n",
" #return l1loss((input.view(-1) + 1) * 112, (targs.view(-1) + 1) * 112);\n",
"\n",
"\n",
"learn = create_cnn(data, models.resnet34, custom_head=head_reg4, metrics=accuracy1, loss_func=loss_func)\n",
"#learn.opt_fn = optim.Adam\n",
"#learn.crit = nn.L1Loss()\n",
"#learn.model\n",
"#print(learn.summary())"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"lr_find(learn)\n",
"learn.recorder.plot()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"scrolled": true
},
"outputs": [
{
"data": {
"text/html": [
"\n",
" <div>\n",
" <style>\n",
" /* Turns off some styling */\n",
" progress {\n",
" /* gets rid of default border in Firefox and Opera. */\n",
" border: none;\n",
" /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
" background-size: auto;\n",
" }\n",
" .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
" background: #F44336;\n",
" }\n",
" </style>\n",
" <progress value='2' class='' max='5', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
" 40.00% [2/5 00:47<01:11]\n",
" </div>\n",
" \n",
"<table style='width:300px; margin-bottom:10px'>\n",
" <tr>\n",
" <th>epoch</th>\n",
" <th>train_loss</th>\n",
" <th>valid_loss</th>\n",
" <th>accuracy1</th>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <th>5.595885</th>\n",
" <th>14.827637</th>\n",
" <th>2016.135132</th>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <th>9.422756</th>\n",
" <th>1.568463</th>\n",
" <th>217.355057</th>\n",
" </tr>\n",
"</table>\n",
"\n",
"\n",
" <div>\n",
" <style>\n",
" /* Turns off some styling */\n",
" progress {\n",
" /* gets rid of default border in Firefox and Opera. */\n",
" border: none;\n",
" /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
" background-size: auto;\n",
" }\n",
" .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
" background: #F44336;\n",
" }\n",
" </style>\n",
" <progress value='8' class='' max='16', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
" 50.00% [8/16 00:06<00:06]\n",
" </div>\n",
" "
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"lr=1e-2\n",
"learn.fit_one_cycle(5, lr)"
]
},
{
"cell_type": "code",
"execution_count": 179,
"metadata": {},
"outputs": [],
"source": [
"learn.save('stage-1')"
]
},
{
"cell_type": "code",
"execution_count": 180,
"metadata": {},
"outputs": [],
"source": [
"learn.load('stage-1')\n",
"lr=5e-3\n",
"learn.freeze_to(-2)"
]
},
{
"cell_type": "code",
"execution_count": 181,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"Total time: 02:04 <p><table style='width:300px; margin-bottom:10px'>\n",
" <tr>\n",
" <th>epoch</th>\n",
" <th>train_loss</th>\n",
" <th>valid_loss</th>\n",
" <th>accuracy1</th>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <th>0.386974</th>\n",
" <th>0.439564</th>\n",
" <th>74.194801</th>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <th>0.389389</th>\n",
" <th>0.996556</th>\n",
" <th>111.589348</th>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <th>0.390334</th>\n",
" <th>1.768569</th>\n",
" <th>146.334457</th>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <th>0.384820</th>\n",
" <th>0.308937</th>\n",
" <th>62.223328</th>\n",
" </tr>\n",
" <tr>\n",
" <th>5</th>\n",
" <th>0.372513</th>\n",
" <th>0.309802</th>\n",
" <th>62.309986</th>\n",
" </tr>\n",
"</table>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"learn.fit_one_cycle(5, slice(lr/10, lr/1))"
]
},
{
"cell_type": "code",
"execution_count": 139,
"metadata": {},
"outputs": [],
"source": [
"learn.save(\"stage-1a\")"
]
},
{
"cell_type": "code",
"execution_count": 64,
"metadata": {
"scrolled": true
},
"outputs": [],
"source": [
"learn.load('mseloss-adamopt-l1crit');"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"scrolled": false
},
"outputs": [],
"source": [
"learn.show_results(rows=10, figsize=(10,40))"
]
},
{
"cell_type": "code",
"execution_count": 107,
"metadata": {},
"outputs": [],
"source": [
"learn.unfreeze()"
]
},
{
"cell_type": "code",
"execution_count": 108,
"metadata": {},
"outputs": [],
"source": [
"lrs = slice(lr/400,lr/4)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"learn.fit_one_cycle(12, lrs)"
]
},
{
"cell_type": "code",
"execution_count": 98,
"metadata": {},
"outputs": [],
"source": [
"learn.save('stage-2');"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Go big - nothing below here is currently in use due to the bad predictions above"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"You may have to restart your kernel and come back to this stage if you run out of memory, and may also need to decrease `bs`."
]
},
{
"cell_type": "code",
"execution_count": 136,
"metadata": {},
"outputs": [],
"source": [
"data = (PointsItemList.from_folder(path_train_sample)\n",
" .random_split_by_pct(0.2)\n",
" .label_from_func(get_y_func)\n",
" .transform(get_transforms(), tfm_y=True, size=256, remove_out=False)\n",
" .databunch(bs=8)\n",
" .normalize(imagenet_stats))"
]
},
{
"cell_type": "code",
"execution_count": 142,
"metadata": {},
"outputs": [],
"source": [
"learn = create_cnn(data, models.resnet50, wd=wd)"
]
},
{
"cell_type": "code",
"execution_count": 143,
"metadata": {},
"outputs": [],
"source": [
"learn.load('stage-2');"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"lr_find(learn)\n",
"learn.recorder.plot()"
]
},
{
"cell_type": "code",
"execution_count": 145,
"metadata": {},
"outputs": [],
"source": [
"lr=1e-4"
]
},
{
"cell_type": "code",
"execution_count": 146,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"Total time: 00:45 <p><table style='width:300px; margin-bottom:10px'>\n",
" <tr>\n",
" <th>epoch</th>\n",
" <th>train_loss</th>\n",
" <th>valid_loss</th>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <th>0.609223</th>\n",
" <th>0.234268</th>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <th>0.612361</th>\n",
" <th>0.225444</th>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <th>0.626123</th>\n",
" <th>0.225538</th>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <th>0.634189</th>\n",
" <th>0.223056</th>\n",
" </tr>\n",
" <tr>\n",
" <th>5</th>\n",
" <th>0.626173</th>\n",
" <th>0.232538</th>\n",
" </tr>\n",
" <tr>\n",
" <th>6</th>\n",
" <th>0.614644</th>\n",
" <th>0.236454</th>\n",
" </tr>\n",
" <tr>\n",
" <th>7</th>\n",
" <th>0.616741</th>\n",
" <th>0.236577</th>\n",
" </tr>\n",
" <tr>\n",
" <th>8</th>\n",
" <th>0.617492</th>\n",
" <th>0.242800</th>\n",
" </tr>\n",
" <tr>\n",
" <th>9</th>\n",
" <th>0.611753</th>\n",
" <th>0.239526</th>\n",
" </tr>\n",
" <tr>\n",
" <th>10</th>\n",
" <th>0.599041</th>\n",
" <th>0.239952</th>\n",
" </tr>\n",
"</table>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"learn.fit_one_cycle(10, slice(lr), pct_start=0.8)"
]
},
{
"cell_type": "code",
"execution_count": 147,
"metadata": {},
"outputs": [],
"source": [
"learn.save('stage-1-big')"
]
},
{
"cell_type": "code",
"execution_count": 148,
"metadata": {},
"outputs": [],
"source": [
"learn.load('stage-1-big');"
]
},
{
"cell_type": "code",
"execution_count": 149,
"metadata": {},
"outputs": [],
"source": [
"learn.unfreeze()"
]
},
{
"cell_type": "code",
"execution_count": 150,
"metadata": {},
"outputs": [],
"source": [
"lrs = slice(1e-6,lr/10)"
]
},
{
"cell_type": "code",
"execution_count": 151,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"Total time: 01:02 <p><table style='width:300px; margin-bottom:10px'>\n",
" <tr>\n",
" <th>epoch</th>\n",
" <th>train_loss</th>\n",
" <th>valid_loss</th>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <th>0.531507</th>\n",
" <th>0.236299</th>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <th>0.555349</th>\n",
" <th>0.236082</th>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <th>0.548737</th>\n",
" <th>0.235379</th>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <th>0.574503</th>\n",
" <th>0.239024</th>\n",
" </tr>\n",
" <tr>\n",
" <th>5</th>\n",
" <th>0.568524</th>\n",
" <th>0.237507</th>\n",
" </tr>\n",
" <tr>\n",
" <th>6</th>\n",
" <th>0.570499</th>\n",
" <th>0.234831</th>\n",
" </tr>\n",
" <tr>\n",
" <th>7</th>\n",
" <th>0.563644</th>\n",
" <th>0.239454</th>\n",
" </tr>\n",
" <tr>\n",
" <th>8</th>\n",
" <th>0.571189</th>\n",
" <th>0.238208</th>\n",
" </tr>\n",
" <tr>\n",
" <th>9</th>\n",
" <th>0.566935</th>\n",
" <th>0.240294</th>\n",
" </tr>\n",
" <tr>\n",
" <th>10</th>\n",
" <th>0.588694</th>\n",
" <th>0.239558</th>\n",
" </tr>\n",
"</table>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"learn.fit_one_cycle(10, lrs)"
]
},
{
"cell_type": "code",
"execution_count": 152,
"metadata": {},
"outputs": [],
"source": [
"learn.save('stage-2-big')"
]
},
{
"cell_type": "code",
"execution_count": 153,
"metadata": {},
"outputs": [],
"source": [
"learn.load('stage-2-big');"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"learn.show_results(rows=3, figsize=(10,10))"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.6.5"
}
},
"nbformat": 4,
"nbformat_minor": 2
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment