I recently perused your GitHub repository dedicated to crowd counting using TensorFlow and I'm immensely impressed with the work you've done. The implementation of a people counter from overhead video surveillance using Transfer Learning is particularly noteworthy.
However, I'd like to raise an issue regarding the model's training dataset. As noted in your README, the model's accuracy is currently limited by the size of the training dataset, which comprises only 30 annotated images. This constraint inevitably affects the model's effectiveness in real-world scenarios, where crowd diversity and density can vary significantly.
To enhance the model's robustness and accuracy, I propose expanding the training dataset. A larger and more varied collection of annotated images would undoubtedly improve the model's ability to accurately count people under different conditions and environments. This expansion could potentially involve crowds of varying sizes, compositions, and settings to create a more comprehensive training regime.
I understand that gathering and annotating such a dataset is a substantial endeavour. However, I believe that the improvement in the model's performance would be well worth the effort, especially for applications in complex surveillance scenarios.
I'm eager to hear your thoughts on this suggestion and would be delighted to contribute to this phase of the project, should you find the proposal beneficial.
Dear Darpan Jain,
I recently perused your GitHub repository dedicated to crowd counting using TensorFlow and I'm immensely impressed with the work you've done. The implementation of a people counter from overhead video surveillance using Transfer Learning is particularly noteworthy.
However, I'd like to raise an issue regarding the model's training dataset. As noted in your README, the model's accuracy is currently limited by the size of the training dataset, which comprises only 30 annotated images. This constraint inevitably affects the model's effectiveness in real-world scenarios, where crowd diversity and density can vary significantly.
To enhance the model's robustness and accuracy, I propose expanding the training dataset. A larger and more varied collection of annotated images would undoubtedly improve the model's ability to accurately count people under different conditions and environments. This expansion could potentially involve crowds of varying sizes, compositions, and settings to create a more comprehensive training regime.
I understand that gathering and annotating such a dataset is a substantial endeavour. However, I believe that the improvement in the model's performance would be well worth the effort, especially for applications in complex surveillance scenarios.
I'm eager to hear your thoughts on this suggestion and would be delighted to contribute to this phase of the project, should you find the proposal beneficial.
Kind regards, yihong1120