# LCFCN

**Repository Path**: deityai/LCFCN

## Basic Information

- **Project Name**: LCFCN
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Apache-2.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-05-11
- **Last Updated**: 2020-12-18

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# LCFCN - ECCV 2018
## Where are the Blobs: Counting by Localization with Point Supervision
[[Paper]](https://arxiv.org/abs/1807.09856)[[Video]](https://youtu.be/DHKD8LGvX6c)

## Requirements

- Pytorch version 0.4 or higher.

## Description
Given a test image, the trained model outputs blobs in the image, then counts the number of predicted blobs (see Figure below).

![Shanghai test image](figures/shanghai.png)

## Test on single image

We test a trained ResNet on a Trancos example image as follows:

```
python main.py -image_path figures/test.png \
                -model_path checkpoints/best_model_trancos_ResFCN.pth \
                -model_name ResFCN
```

The expected output is shown below, and the output image will be saved in the same directory as the test image.

Trancos test image           |  Trancos predicted image
:-------------------------:|:-------------------------:
![Trancos test image](figures/test.png) |  ![Trancos pred image](figures/test.png_blobs_count:32.png)


## Running the saved models

1. Download the checkpoints,
```
bash checkpoints/download.sh
```
For the shanghai model, download the checkpoint from this link:

https://drive.google.com/file/d/1N75fun1I1XWh1LuKmi60QXF2SgCPLLLQ/view?usp=sharing

2. Output the saved results,

```
python main.py -m summary -e trancos
```

3. Re-evaluate the saved model,

```
python main.py -m test -e trancos
```


## Training the models from scratch

To train the model,

```
python main.py -m train -e trancos
```


## Benchmark

| Method           | Trancos | Pascal|
|------------------|---------|-------|
| ResFCN           | 3.39    | 0.31   |  
| Paper            | 3.32    | 0.31   | 




## Citation 
If you find the code useful for your research, please cite:

```bibtex
@inproceedings{laradji2018blobs,
  title={Where are the blobs: Counting by localization with point supervision},
  author={Laradji, Issam H and Rostamzadeh, Negar and Pinheiro, Pedro O and Vazquez, David and Schmidt, Mark},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  pages={547--562},
  year={2018}
}
```