analyze_3d_model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from ucsgnet.ucsgnet.net_3d import Net\n",
    "\n",
    "import torch\n",
    "import shlex\n",
    "import argparse\n",
    "import os\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from tqdm.auto import tqdm\n",
    "from sklearn.manifold import TSNE\n",
    "from collections import defaultdict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_args() -> argparse.Namespace:\n",
    "    parser = argparse.ArgumentParser(\n",
    "        description=(\n",
    "            \"Reconstructs all shapes in the dataset by predicting values at \"\n",
    "            \"each 3D point and then thresholding\"\n",
    "        ),\n",
    "        add_help=False,\n",
    "    )\n",
    "\n",
    "    parser.add_argument(\n",
    "        \"--weights_path\", required=True, help=\"Path to the model to load\"\n",
    "    )\n",
    "    parser.add_argument(\n",
    "        \"--size\", type=int, help=\"Data size to be used\", required=True\n",
    "    )\n",
    "    parser.add_argument(\n",
    "        \"--processed\",\n",
    "        dest=\"processed_data_path\",\n",
    "        type=str,\n",
    "        help=\"Base folder of processed data\",\n",
    "        required=True,\n",
    "    )\n",
    "    parser.add_argument(\n",
    "        \"--valid\",\n",
    "        dest=\"valid_file\",\n",
    "        type=str,\n",
    "        help=\"Path to valid HDF5 file with the valid data\",\n",
    "        required=True,\n",
    "    )\n",
    "    parser.add_argument(\n",
    "        \"--valid_shape_names\",\n",
    "        type=str,\n",
    "        help=(\n",
    "            \"Path to valid text file with the names for each data point in \"\n",
    "            \"the valid dataset\"\n",
    "        ),\n",
    "        required=True,\n",
    "    )\n",
    "    parser.add_argument(\n",
    "        \"--sphere_complexity\",\n",
    "        type=int,\n",
    "        help=\"Number of segments lat/lon of the sphere\",\n",
    "        required=False,\n",
    "        default=16,\n",
    "    )\n",
    "\n",
    "    parser = Net.add_model_specific_args(parser)\n",
    "    args = parser.parse_args(\n",
    "        shlex.split(\n",
    "            \"--valid all_vox256_img_test.hdf5 --valid_shape_names all_vox256_img_test.txt --processed data/hdf5 --size 64 --weights_path models/3d_64/initial/ckpts/model.ckpt --sphere_complexity 3\"\n",
    "        )\n",
    "    )\n",
    "    return args\n",
    "\n",
    "args = get_args()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Num of trainable params: 16641062\n",
      "Num of not trainable params: 0\n"
     ]
    }
   ],
   "source": [
    "model = Net.load_from_checkpoint(args.weights_path)\n",
    "model.build(\"\", args.valid_file, args.processed_data_path, 64)\n",
    "model.turn_fine_tuning_mode()\n",
    "model.freeze()\n",
    "model.hparams.batch_size = 12\n",
    "if torch.cuda.is_available():\n",
    "    model = model.cuda()\n",
    "with open(\n",
    "    os.path.join(args.processed_data_path, args.valid_shape_names)\n",
    ") as f:\n",
    "    file_names = f.read().split(\"\\n\")\n",
    "\n",
    "loader = model.val_dataloader()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7c98d74d722d4786908fd92944862670",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=731.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "device = next(model.parameters()).device\n",
    "latents = []\n",
    "latent_codes_for_samples = defaultdict(list)\n",
    "emits = defaultdict(list)\n",
    "for _, batch in enumerate(tqdm(loader)):\n",
    "    voxels = batch[0].to(device)\n",
    "    points = batch[1].to(device)\n",
    "    vox_pred = (\n",
    "        model.net.encoder_(voxels)\n",
    "    )\n",
    "    \n",
    "    latent_codes, batch_emits = model.net.get_latent_codes_for_each_layer(\n",
    "        voxels, points\n",
    "    )\n",
    "    for key, elems in latent_codes.items():\n",
    "        latent_codes_for_samples[key].append(elems)\n",
    "        \n",
    "    for key, elems in batch_emits.items():\n",
    "        emits[key].append(elems)\n",
    "    latents.append(vox_pred)\n",
    "latents = torch.cat(latents, dim=0).detach().cpu().numpy()\n",
    "latent_codes_for_samples = {key: torch.cat(lats, dim=0).detach().cpu().numpy() for key, lats in latent_codes_for_samples.items()}\n",
    "emits = {key: torch.cat(lats, dim=0).detach().cpu().numpy() for key, lats in emits.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(latents.reshape((-1,)), bins=1000)\n",
    "print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(model.net._gru_hidden_state.detach().cpu().numpy()[0], bins=100)\n",
    "print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(emits['emits_3'].reshape((-1,)), bins=1000)\n",
    "print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "tsne = TSemitscomponents=2, verbose=1, perplexity=40, n_iter=300)\n",
    "tsne_results = tsne.fit_transform(latents)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "cat_to_str = {name.split(\"_\")[0]: name.split(\"_\")[1] for name in [\n",
    "    \"02691156_airplane\",\n",
    "    \"02828884_bench\",\n",
    "    \"02933112_cabinet\",\n",
    "    \"02958343_car\",\n",
    "    \"03001627_chair\",\n",
    "    \"03211117_display\",\n",
    "    \"03636649_lamp\",\n",
    "    \"03691459_speaker\",\n",
    "    \"04090263_rifle\",\n",
    "    \"04256520_couch\",\n",
    "    \"04379243_table\",\n",
    "    \"04401088_phone\",\n",
    "    \"04530566_vessel\",\n",
    "]}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = {}\n",
    "data['y'] = tsne_results[:,0]\n",
    "data['x'] = tsne_results[:,1]\n",
    "classes = [cat_to_str[name.split('/')[0]] for name in file_names[:-1]]\n",
    "data[\"Object\"] = classes\n",
    "data = pd.DataFrame(data)\n",
    "plt.figure(figsize=(16,10))\n",
    "ax = sns.scatterplot(\n",
    "    x=\"x\", y=\"y\",\n",
    "    hue=\"Object\",\n",
    "    palette=sns.color_palette(\"hls\", 13),\n",
    "    data=data,\n",
    "    legend=\"full\",\n",
    "    alpha=0.3\n",
    ")\n",
    "ax.set_ylabel(\"off\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data['y']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}