Update README.md

README.md

@@ -12,7 +12,7 @@ license: mit
 
 # ERA-SESSION13 YoloV3 with Pytorch Lightning & Gradio
 
-HF Link: https://huggingface.co/spaces/RaviNaik/ERA-SESSION13
+HF Link: https://huggingface.co/spaces/GunaKoppula/Session13
 
 ### Achieved:
 1. **Training Loss: 3.680** 
@@ -21,13 +21,42 @@ HF Link: https://huggingface.co/spaces/RaviNaik/ERA-SESSION13
 4. **No obj accuracy: 97.991463%** 
 5. **Obj accuracy: 75.976616%**
 6. **MAP: 0.4366795**
+
+### Tasks:
+1. :heavy_check_mark: Move the code to PytorchLightning
+2. :heavy_check_mark: Train the model to reach such that all of these are true:
+    - Class accuracy is more than 75%
+    - No Obj accuracy of more than 95%
+    - Object Accuracy of more than 70% (assuming you had to reduce the kernel numbers, else 80/98/78)
+    - Ideally trained till 40 epochs
+3. :heavy_check_mark: Add these training features:
+    - Add multi-resolution training - the code shared trains only on one resolution 416
+    - Add Implement Mosaic Augmentation only 75% of the times
+    - Train on float16
+    - GradCam must be implemented.
+4. :heavy_check_mark: Things that are allowed due to HW constraints:
+    - Change of batch size
+    - Change of resolution
+    - Change of OCP parameters
+5. :heavy_check_mark: Once done:
+    - Move the app to HuggingFace Spaces
+    - Allow custom upload of images
+    - Share some samples from the existing dataset
+    - Show the GradCAM output for the image that the user uploads as well as for the samples.
+6. :heavy_check_mark: Mention things like:
+    - classes that your model support
+    - link to the actual model
+7. :heavy_check_mark: Assignment:
+    - Share HuggingFace App Link
+    - Share LightningCode Link on Github
+    - Share notebook link (with logs) on GitHub
   
 ### Results
-![image](https://github.com/RaviNaik/ERA-SESSION13/blob/main/yolo_results.png)
+![image](https://github.com/GunaKoppula/ERAV1-Session-13/blob/main/yolo_results.png)
 
 ### Gradio App
-![image](https://github.com/RaviNaik/ERA-SESSION13/assets/23289802/95335687-e717-4467-bcb1-227a79dd5c3f)
-![image](https://github.com/RaviNaik/ERA-SESSION13/assets/23289802/3ab67d32-38e6-436a-86d4-b76b5bd52a77)
+![image](https://github.com/GunaKoppula/ERAV1-Session-13/assets/61241928/5304d1e4-a545-4b8c-951c-10cc5da09e00)
+![image](https://github.com/GunaKoppula/ERAV1-Session-13/assets/61241928/8d558059-4477-4d54-a0c6-9f6bb424c77c)
 
 ### Model Summary
 ```python
@@ -397,11 +426,11 @@ HF Link: https://huggingface.co/spaces/RaviNaik/ERA-SESSION13
 ```
 
 ### LR Finder
-![image](https://github.com/RaviNaik/ERA-SESSION13/assets/23289802/a6d64f13-a7b7-4e17-abfc-3ec86e84b710)
+![image](https://github.com/GunaKoppula/ERAV1-Session-13/assets/61241928/7ffabc81-f1d3-4379-bbfb-6ba7da277a02)
 
 ### Loss & Accuracy
 **Training & Validation Loss:**
-![image](https://github.com/RaviNaik/ERA-SESSION13/assets/23289802/9391157e-a889-480d-b233-b72e86745245)
+![image](https://github.com/GunaKoppula/ERAV1-Session-13/assets/61241928/332fda1e-acfb-4aec-979f-93984bc43e2d)
 
 **Testing Accuracy:**
 ```python
@@ -494,12 +523,15 @@ Obj accuracy is: 75.976616%
 MAP: 0.43667954206466675
 ```
 ### Tensorboard Plots 
-**Training Loss vs Steps:** ![image](https://github.com/RaviNaik/ERA-SESSION13/assets/23289802/5cb753e0-377b-4d9f-a240-871270ed50db) 
+**Training Loss vs Steps:** ![image](https://github.com/GunaKoppula/ERAV1-Session-13/assets/61241928/3b4fb334-5b2a-45b0-a892-5222c147160a)
 
 **Validation Loss vs Steps:** 
-(Info: Validation loss calculated every 10 epochs to save time, thats why the straight line) 
-![image](https://github.com/RaviNaik/ERA-SESSION13/assets/23289802/7401c0aa-f7ff-4a5b-bab2-dbb5ebe0b400) 
+(Info: Validation loss calculated every 10 epochs to save time, that's why the straight line) 
+![image](https://github.com/GunaKoppula/ERAV1-Session-13/assets/61241928/8b5f2b66-41d9-40bc-80f9-94589d5ffb59)
 
 ### GradCAM Representations
-EigenCAM is used to generate CAM representation, since usal gradient based method wont work with detection models like Yolo, FRCNN etc.
-![image](https://github.com/RaviNaik/ERA-SESSION13/assets/23289802/3e3917f1-c8d1-4c3f-a028-de1292575e0b)
+EigenCAM is used to generate CAM representation, since the usual gradient-based method won't work with detection models like Yolo, FRCNN, etc.
+![image](https://github.com/GunaKoppula/ERAV1-Session-13/assets/61241928/0f679781-fe30-41f8-9625-8fa312ae7f38)
+
+
+