Upgrade

Author: varunagrawal

Makefile

@@ -23,30 +23,30 @@ CHECKPOINT=weights/vqa_checkpoint_DeeperLSTM_200.pth
 main: train evaluate
 
 train:
-    python main.py $(TRAIN_ANN) $(TRAIN_QUES) $(VAL_ANN) $(VAL_QUES) --images $(TRAIN_IMGS) --val_images $(VAL_IMGS) \
+	uv run python main.py $(TRAIN_ANN) $(TRAIN_QUES) $(VAL_ANN) $(VAL_QUES) --images $(TRAIN_IMGS) --val_images $(VAL_IMGS) \
     --arch $(ARCH) --batch_size ${BATCH} --num_workers ${WORKERS} --image_root $(IMAGE_ROOT) --image_dim $(IMAGE_DIM)
 
 train_mcb:
-    python main.py $(TRAIN_ANN) $(TRAIN_QUES) $(VAL_ANN) $(VAL_QUES) --images $(TRAIN_IMGS) --val_images $(VAL_IMGS) \
+	uv run python main.py $(TRAIN_ANN) $(TRAIN_QUES) $(VAL_ANN) $(VAL_QUES) --images $(TRAIN_IMGS) --val_images $(VAL_IMGS) \
     --arch MCBModel --batch_size 64 --num_workers ${WORKERS} --image_root $(IMAGE_ROOT) --raw_images --img_size 448 --image_dim $(IMAGE_DIM)
 
 
 raw_images:
-    python main.py $(TRAIN_ANN) $(TRAIN_QUES) $(VAL_ANN) $(VAL_QUES) \
+	uv run python main.py $(TRAIN_ANN) $(TRAIN_QUES) $(VAL_ANN) $(VAL_QUES) \
     --raw_images --image_root $(IMAGE_ROOT) --arch $(ARCH) --batch_size 32
 
 options:
-    python main.py -h
+	python main.py -h
 
 evaluate:
-    python evaluate.py $(TRAIN_ANN) $(TRAIN_QUES) $(VAL_ANN) $(VAL_QUES) --images $(TRAIN_IMGS) --val_images $(VAL_IMGS) \
-      --batch_size 1 --resume $(CHECKPOINT) --num_workers ${WORKERS} --image_dim $(IMAGE_DIM)
+	python evaluate.py $(TRAIN_ANN) $(TRAIN_QUES) $(VAL_ANN) $(VAL_QUES) --images $(TRAIN_IMGS) --val_images $(VAL_IMGS) \
+    --batch_size 1 --resume $(CHECKPOINT) --num_workers ${WORKERS} --image_dim $(IMAGE_DIM)
 
 preprocess_train:
-    python preprocess_images.py $(IMAGE_ROOT)/annotations/instances_train2014.json --root $(IMAGE_ROOT) --split train --arch resnet152
+	uv run python preprocess_images.py $(IMAGE_ROOT)/annotations/instances_train2014.json --root $(IMAGE_ROOT) --split train --arch resnet152
 
 preprocess_val:
-    python preprocess_images.py $(IMAGE_ROOT)/annotations/instances_val2014.json --root $(IMAGE_ROOT) --split val --arch resnet152
+	python preprocess_images.py $(IMAGE_ROOT)/annotations/instances_val2014.json --root $(IMAGE_ROOT) --split val --arch resnet152
 
 demo:
-    python demo.py demo_img.jpg "what room is this?" $(TRAIN_QUES) $(TRAIN_ANN) --checkpoint $(CHECKPOINT)
+	python demo.py demo_img.jpg "what room is this?" $(TRAIN_QUES) $(TRAIN_ANN) --checkpoint $(CHECKPOINT)

README.md

@@ -2,7 +2,6 @@
 
 Visual Question Answering in PyTorch
 
-
 ## Setup
 
 - Run `pip install -r requirements.txt` to install all the required python packages.
@@ -11,36 +10,40 @@ Visual Question Answering in PyTorch
 ## Preprocess Images
 
 We assume you use image embeddings, which you can process using `preprocess_images.py`.
+
 ```shell
 python preprocess_images.py <path to instances_train2014.json> \
     --root <path to dataset root "train2014|val2014"> \
     --split <train|val> --arch <vgg16|vgg19_bn|resnet152>
 ```
+
 I have already pre-processed all the COCO images (both train and test sets) using the VGG-16, VGG-19-BN, and ResNet-152 models. To download them, please go into the `image_embeddings` directory and run `make <model>`.</br>
 Here `<model>` can be either `vgg16`, `vgg19_bn` or `resnet152` depending on which model's embeddings you need. E.g. `make resnet152`
 
 Alternatively, you can find them [here](https://1drv.ms/f/s!Au18pri6pxSNlop81AhX4bATqy1VJA).
 
-
 ## Running
 
 ### Training
 
 To run the training and evaluation code with default values, just type
 
-```
+```shell
 make
 ```
 
 If you wish to only run the training code, you can run
-```
+
+```shell
 make train
 ```
 
 If you want to use the raw RGB images from COCO, you can type
+
 ```shell
 make raw_images
 ```
+
 This takes the same arguments as `make train`.
 
 You can get a list of options with `make options` or `python main.py -h`.
@@ -49,7 +52,6 @@ Check out the `Makefile` to get an idea of how to run the code.
 
 > *NOTE* The code will take care of all the text preprocessing. Just sit back and relax.
 
-
 The minimum arguments required are:
 
 1. The VQA train annotations dataset
@@ -61,15 +63,15 @@ The minimum arguments required are:
 
 ### Evaluation
 
-Evaluating the performance of the model on a fine-grained basis is important. Thus this repo supports evaluating 
-answers to questions based on answer type (e.g. "yes/no" questions).
+Evaluating the performance of the model on a fine-grained basis is important. Thus this repo supports evaluating  answers to questions based on answer type (e.g. "yes/no" questions).
 
 To evaluate the model, run
+
 ```shell
 make evaluate
 ```
 
-You are required to pass in the `--resume` argument to point to the trained model weights. The other arguments are 
+You are required to pass in the `--resume` argument to point to the trained model weights. The other arguments are
 the same as in training.
 
 ### Demo
@@ -86,7 +88,6 @@ You can use your own image or question:
 python demo.py demo_img.jpg "what room is this?"
 ```
 
-
 ## Results
 
 **NOTE** We train and evaluate on the balanced datasets.

demo.py

@@ -1,25 +1,25 @@
 import argparse
-from models import Models
-from utils import image, text
-from PIL import Image, ImageFont, ImageDraw
-import numpy as np
+
 import torch
-from torch import nn
+from PIL import Image, ImageDraw,
 from torchvision import transforms
+
 from dataset import process_vqa_dataset
+from models import Models
+from utils import image, text
 
 
 def parse_args():
     parser = argparse.ArgumentParser("VQA Demo")
     parser.add_argument("image", help="Path to image file")
     parser.add_argument("question", help="Question text")
-    parser.add_argument(
-        "questions", help="Path to VQA Questions training file")
-    parser.add_argument(
-        "annotations", help="Path to COCO Annotations training file")
+    parser.add_argument("questions",
+                        help="Path to VQA Questions training file")
+    parser.add_argument("annotations",
+                        help="Path to COCO Annotations training file")
     parser.add_argument("--model", default="DeeperLSTM")
-    parser.add_argument(
-        "--checkpoint", default="weights/deeper_lstm_best_weights.pth.tar")
+    parser.add_argument("--checkpoint",
+                        default="weights/deeper_lstm_best_weights.pth.tar")
     parser.add_argument("--preprocessed_cache",
                         default="vqa_train_dataset_cache.pickle")
     parser.add_argument("--embedding_arch", default="vgg19_bn")
@@ -66,8 +66,10 @@ def main():
 
     try:
         weights = torch.load(args.checkpoint)
-    except (Exception,):
-        print("ERROR: Default weights missing. Please specify weights for the VQA model")
+    except (Exception, ):
+        print(
+            "ERROR: Default weights missing. Please specify weights for the VQA model"
+        )
         exit(0)
 
     model.load_state_dict(weights["model"])

evaluate.py

@@ -1,13 +1,13 @@
+import json
 import os.path as osp
-import dataset
-from models.model import Models
-import numpy as np
+
 import torch
-from torch.nn import functional
 from torchvision import transforms
-from arguments import parse_args
 from tqdm import tqdm
-import json
+
+import dataset
+from arguments import parse_args
+from models.model import Models
 
 
 @torch.no_grad()

metrics.py

@@ -1,5 +1,9 @@
+"""Metrics module."""
+
+
 def accuracy(output, target):
-    pred, indices = output.max(dim=1)
+    """Compute the accuracy between the `output` and the `target`."""
+    _, indices = output.max(dim=1)
     acc = indices.eq(target).float().sum()
     acc = acc / target.size(0)
     return acc

pyproject.toml

@@ -0,0 +1,35 @@
+[project]
+name = "VisualQA"
+version = "1.0.0" 
+
+description = "Visual Question Answering in PyTorch"
+readme = "README.md"
+authors = [
+    { name = "Varun Agrawal", email = "varagrawal@gmail.com" }
+]
+requires-python = ">= 3.11"
+dependencies = [
+  "nltk>=3.9.2",
+  "numpy>=2.0.0",
+  "pillow>=12.0.0",
+  "torch>=2.6.0",
+  "torchvision>=0.24",
+  "tqdm>=4.67.1",
+  "visdom>=0.2.4",
+]
+
+classifiers = [
+    "Operating System :: MacOS",
+    "Operating System :: Microsoft :: Windows",
+    "Operating System :: POSIX :: Linux",
+    "Programming Language :: Python",
+    "Programming Language :: Python :: 3.11",
+]
+
+[project.license]
+text = "MIT"
+
+[project.urls]
+Homepage = "https://github.com/varunagrawal/VisualQA"
+Repository = "https://github.com/varunagrawal/VisualQA"
+