Visual Question Answering on CLEVR-Humans (test)
81.7AccuracyMDETR
Evaluation Results
| Method | Links | |
|---|---|---|
| MDETRfine-tuned=true2021.04 | 81.7 | |
| MACfine-tuned=true2021.04 | 81.5 | |
| CNN+GRU+FiLMTraining Protocol=Train CLEVR, fine-tune human2017.09 | 75.9 | |
| FiLMfine-tuned=true2021.04 | 75.9 | |
| NS-VQAfine-tuned=false, external program annotations=true2021.04 | 67.8 | |
| PG+EE (18K prog.)Training Protocol=Train CLEVR, fine-tune human2017.09 | 66.6 | |
| MDETRfine-tuned=false2021.04 | 59.9 | |
| CNN+LSTM+SA+MLPTraining Protocol=Train CLEVR, fine-tune human2017.09 | 57.6 | |
| MACfine-tuned=false2021.04 | 57.4 | |
| CNN+GRU+FiLMTraining Protocol=Train CLEVR2017.09 | 56.6 | |
| FiLMfine-tuned=false2021.04 | 56.6 | |
| PG+EE (18K prog.)Training Protocol=Train CLEVR2017.09 | 54 | |
| CNN+LSTM+SA+MLPTraining Protocol=Train CLEVR2017.09 | 50.4 | |
| CNN+LSTMTraining Protocol=Train CLEVR, fine-tune human2017.09 | 43.2 | |
| CNN+LSTMTraining Protocol=Train CLEVR2017.09 | 37.7 | |
| LSTMTraining Protocol=Train CLEVR, fine-tune human2017.09 | 36.5 | |
| LSTMTraining Protocol=Train CLEVR2017.09 | 27.5 |