Pytorch: Inspect Activations Layer-by-layer in Trained Model

inspect-trained-model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torchvision import models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Model(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        net = models.vgg11(pretrained=True)\n",
    "        self.features = nn.Sequential(*list(net.features))\n",
    "        self.classifier = nn.Sequential(*list(net.classifier))\n",
    "        \n",
    "    def forward(self, x):\n",
    "        outputs = []\n",
    "        \n",
    "        for layer in self.features:\n",
    "            x = layer(x)\n",
    "            outputs.append(x)\n",
    "        \n",
    "        # flatten\n",
    "        x = x.view(x.size(0), -1)\n",
    "        \n",
    "        for layer in self.classifier:\n",
    "            x = layer(x)\n",
    "            outputs.append(x)\n",
    "        \n",
    "        return x, outputs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "m = Model().cuda()\n",
    "x = torch.randn(10, 3, 224, 224).cuda()\n",
    "x, activations = m(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([10, 1000])\n"
     ]
    }
   ],
   "source": [
    "print(x.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([10, 64, 224, 224])\n",
      "torch.Size([10, 64, 224, 224])\n",
      "torch.Size([10, 64, 112, 112])\n",
      "torch.Size([10, 128, 112, 112])\n",
      "torch.Size([10, 128, 112, 112])\n",
      "torch.Size([10, 128, 56, 56])\n",
      "torch.Size([10, 256, 56, 56])\n",
      "torch.Size([10, 256, 56, 56])\n",
      "torch.Size([10, 256, 56, 56])\n",
      "torch.Size([10, 256, 56, 56])\n",
      "torch.Size([10, 256, 28, 28])\n",
      "torch.Size([10, 512, 28, 28])\n",
      "torch.Size([10, 512, 28, 28])\n",
      "torch.Size([10, 512, 28, 28])\n",
      "torch.Size([10, 512, 28, 28])\n",
      "torch.Size([10, 512, 14, 14])\n",
      "torch.Size([10, 512, 14, 14])\n",
      "torch.Size([10, 512, 14, 14])\n",
      "torch.Size([10, 512, 14, 14])\n",
      "torch.Size([10, 512, 14, 14])\n",
      "torch.Size([10, 512, 7, 7])\n",
      "torch.Size([10, 4096])\n",
      "torch.Size([10, 4096])\n",
      "torch.Size([10, 4096])\n",
      "torch.Size([10, 4096])\n",
      "torch.Size([10, 4096])\n",
      "torch.Size([10, 4096])\n",
      "torch.Size([10, 1000])\n"
     ]
    }
   ],
   "source": [
    "for a in activations:\n",
    "    print(a.shape)"
   ]
  },
  {
   "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": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}